blob: 17c00d93de2e4973278845e98554ce2040445f65 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/nommu.c
3 *
4 * Replacement code for mm functions to support CPU's that don't
5 * have any form of memory management unit (thus no virtual memory).
6 *
7 * See Documentation/nommu-mmap.txt
8 *
David Howells8feae132009-01-08 12:04:47 +00009 * Copyright (c) 2004-2008 David Howells <dhowells@redhat.com>
Linus Torvalds1da177e2005-04-16 15:20:36 -070010 * Copyright (c) 2000-2003 David McCullough <davidm@snapgear.com>
11 * Copyright (c) 2000-2001 D Jeff Dionne <jeff@uClinux.org>
12 * Copyright (c) 2002 Greg Ungerer <gerg@snapgear.com>
Paul Mundt29c185e2010-12-24 12:08:30 +090013 * Copyright (c) 2007-2010 Paul Mundt <lethal@linux-sh.org>
Linus Torvalds1da177e2005-04-16 15:20:36 -070014 */
15
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -070016#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
17
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040018#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include <linux/mm.h>
Ingo Molnar6e84f312017-02-08 18:51:29 +010020#include <linux/sched/mm.h>
Davidlohr Bueso615d6e82014-04-07 15:37:25 -070021#include <linux/vmacache.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/mman.h>
23#include <linux/swap.h>
24#include <linux/file.h>
25#include <linux/highmem.h>
26#include <linux/pagemap.h>
27#include <linux/slab.h>
28#include <linux/vmalloc.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <linux/blkdev.h>
30#include <linux/backing-dev.h>
Gideon Israel Dsouza3b321232014-04-07 15:37:26 -070031#include <linux/compiler.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include <linux/mount.h>
33#include <linux/personality.h>
34#include <linux/security.h>
35#include <linux/syscalls.h>
Al Viro120a7952010-10-30 02:54:44 -040036#include <linux/audit.h>
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -070037#include <linux/printk.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -080039#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070040#include <asm/tlb.h>
41#include <asm/tlbflush.h>
Bernd Schmidteb8cdec2009-09-21 17:03:57 -070042#include <asm/mmu_context.h>
David Howells8feae132009-01-08 12:04:47 +000043#include "internal.h"
44
Linus Torvalds1da177e2005-04-16 15:20:36 -070045void *high_memory;
Arnd Bergmann944b6872015-02-05 12:25:12 -080046EXPORT_SYMBOL(high_memory);
Linus Torvalds1da177e2005-04-16 15:20:36 -070047struct page *mem_map;
48unsigned long max_mapnr;
gchen gchen5b8bf302015-03-12 16:26:05 -070049EXPORT_SYMBOL(max_mapnr);
Hugh Dickins4266c972009-09-23 17:05:53 +010050unsigned long highest_memmap_pfn;
David Howellsfc4d5c22009-05-06 16:03:05 -070051int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -070052int heap_stack_gap = 0;
53
David Howells33e5d7692009-04-02 16:56:32 -070054atomic_long_t mmap_pages_allocated;
David Howells8feae132009-01-08 12:04:47 +000055
Linus Torvalds1da177e2005-04-16 15:20:36 -070056EXPORT_SYMBOL(mem_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -070057
David Howells8feae132009-01-08 12:04:47 +000058/* list of mapped, potentially shareable regions */
59static struct kmem_cache *vm_region_jar;
60struct rb_root nommu_region_tree = RB_ROOT;
61DECLARE_RWSEM(nommu_region_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -070062
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +040063const struct vm_operations_struct generic_file_vm_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -070064};
65
66/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070067 * Return the total memory allocated for this pointer, not
68 * just what the caller asked for.
69 *
70 * Doesn't have to be accurate, i.e. may have races.
71 */
72unsigned int kobjsize(const void *objp)
73{
74 struct page *page;
75
Michael Hennerich4016a132008-04-28 02:13:38 -070076 /*
77 * If the object we have should not have ksize performed on it,
78 * return size of 0
79 */
Paul Mundt5a1603b2008-06-12 16:29:55 +090080 if (!objp || !virt_addr_valid(objp))
Paul Mundt6cfd53fc2008-06-05 22:46:08 -070081 return 0;
82
83 page = virt_to_head_page(objp);
Paul Mundt6cfd53fc2008-06-05 22:46:08 -070084
85 /*
86 * If the allocator sets PageSlab, we know the pointer came from
87 * kmalloc().
88 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070089 if (PageSlab(page))
90 return ksize(objp);
91
Paul Mundt6cfd53fc2008-06-05 22:46:08 -070092 /*
Paul Mundtab2e83e2009-01-08 12:04:48 +000093 * If it's not a compound page, see if we have a matching VMA
94 * region. This test is intentionally done in reverse order,
95 * so if there's no VMA, we still fall through and hand back
96 * PAGE_SIZE for 0-order pages.
97 */
98 if (!PageCompound(page)) {
99 struct vm_area_struct *vma;
100
101 vma = find_vma(current->mm, (unsigned long)objp);
102 if (vma)
103 return vma->vm_end - vma->vm_start;
104 }
105
106 /*
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700107 * The ksize() function is only guaranteed to work for pointers
Paul Mundt5a1603b2008-06-12 16:29:55 +0900108 * returned by kmalloc(). So handle arbitrary pointers here.
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700109 */
Paul Mundt5a1603b2008-06-12 16:29:55 +0900110 return PAGE_SIZE << compound_order(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111}
112
Lorenzo Stoakes0d731752016-10-24 10:57:25 +0100113static long __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
Michel Lespinasse28a35712013-02-22 16:35:55 -0800114 unsigned long start, unsigned long nr_pages,
115 unsigned int foll_flags, struct page **pages,
116 struct vm_area_struct **vmas, int *nonblocking)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117{
Sonic Zhang910e46d2006-09-27 01:50:17 -0700118 struct vm_area_struct *vma;
David Howells7b4d5b82006-09-27 01:50:18 -0700119 unsigned long vm_flags;
120 int i;
121
122 /* calculate required read or write permissions.
Hugh Dickins58fa8792009-09-21 17:03:31 -0700123 * If FOLL_FORCE is set, we only require the "MAY" flags.
David Howells7b4d5b82006-09-27 01:50:18 -0700124 */
Hugh Dickins58fa8792009-09-21 17:03:31 -0700125 vm_flags = (foll_flags & FOLL_WRITE) ?
126 (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
127 vm_flags &= (foll_flags & FOLL_FORCE) ?
128 (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129
Peter Zijlstra9d737772009-06-25 11:58:55 +0200130 for (i = 0; i < nr_pages; i++) {
David Howells7561e8c2010-03-25 16:48:38 +0000131 vma = find_vma(mm, start);
David Howells7b4d5b82006-09-27 01:50:18 -0700132 if (!vma)
133 goto finish_or_fault;
134
135 /* protect what we can, including chardevs */
Hugh Dickins1c3aff12009-09-21 17:03:24 -0700136 if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
137 !(vm_flags & vma->vm_flags))
David Howells7b4d5b82006-09-27 01:50:18 -0700138 goto finish_or_fault;
Sonic Zhang910e46d2006-09-27 01:50:17 -0700139
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140 if (pages) {
141 pages[i] = virt_to_page(start);
142 if (pages[i])
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300143 get_page(pages[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 }
145 if (vmas)
Sonic Zhang910e46d2006-09-27 01:50:17 -0700146 vmas[i] = vma;
David Howellse1ee65d2010-03-25 16:48:44 +0000147 start = (start + PAGE_SIZE) & PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148 }
David Howells7b4d5b82006-09-27 01:50:18 -0700149
150 return i;
151
152finish_or_fault:
153 return i ? : -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154}
Nick Pigginb291f002008-10-18 20:26:44 -0700155
Nick Pigginb291f002008-10-18 20:26:44 -0700156/*
157 * get a list of pages in an address range belonging to the specified process
158 * and indicate the VMA that covers each page
159 * - this is potentially dodgy as we may end incrementing the page count of a
160 * slab page or a secondary page from a compound page
161 * - don't permit access to VMAs that don't support it, such as I/O mappings
162 */
Ingo Molnarc12d2da2016-04-04 10:24:58 +0200163long get_user_pages(unsigned long start, unsigned long nr_pages,
Lorenzo Stoakes768ae302016-10-13 01:20:16 +0100164 unsigned int gup_flags, struct page **pages,
Michel Lespinasse28a35712013-02-22 16:35:55 -0800165 struct vm_area_struct **vmas)
Nick Pigginb291f002008-10-18 20:26:44 -0700166{
Lorenzo Stoakes768ae302016-10-13 01:20:16 +0100167 return __get_user_pages(current, current->mm, start, nr_pages,
168 gup_flags, pages, vmas, NULL);
Nick Pigginb291f002008-10-18 20:26:44 -0700169}
Ingo Molnarc12d2da2016-04-04 10:24:58 +0200170EXPORT_SYMBOL(get_user_pages);
Greg Ungerer66aa2b42005-09-12 11:18:10 +1000171
Ingo Molnarc12d2da2016-04-04 10:24:58 +0200172long get_user_pages_locked(unsigned long start, unsigned long nr_pages,
Lorenzo Stoakes3b913172016-10-13 01:20:14 +0100173 unsigned int gup_flags, struct page **pages,
Dave Hansencde70142016-02-12 13:01:55 -0800174 int *locked)
Andrea Arcangelif0818f42015-02-11 15:27:17 -0800175{
Lorenzo Stoakes768ae302016-10-13 01:20:16 +0100176 return get_user_pages(start, nr_pages, gup_flags, pages, NULL);
Andrea Arcangelif0818f42015-02-11 15:27:17 -0800177}
Ingo Molnarc12d2da2016-04-04 10:24:58 +0200178EXPORT_SYMBOL(get_user_pages_locked);
Andrea Arcangelif0818f42015-02-11 15:27:17 -0800179
Lorenzo Stoakes8b7457e2016-12-14 15:06:55 -0800180static long __get_user_pages_unlocked(struct task_struct *tsk,
181 struct mm_struct *mm, unsigned long start,
182 unsigned long nr_pages, struct page **pages,
183 unsigned int gup_flags)
Andrea Arcangelif0818f42015-02-11 15:27:17 -0800184{
185 long ret;
186 down_read(&mm->mmap_sem);
Dave Hansencde70142016-02-12 13:01:55 -0800187 ret = __get_user_pages(tsk, mm, start, nr_pages, gup_flags, pages,
188 NULL, NULL);
Andrea Arcangelif0818f42015-02-11 15:27:17 -0800189 up_read(&mm->mmap_sem);
190 return ret;
191}
Andrea Arcangeli0fd71a52015-02-11 15:27:20 -0800192
Ingo Molnarc12d2da2016-04-04 10:24:58 +0200193long get_user_pages_unlocked(unsigned long start, unsigned long nr_pages,
Lorenzo Stoakesc1641542016-10-13 01:20:13 +0100194 struct page **pages, unsigned int gup_flags)
Andrea Arcangeli0fd71a52015-02-11 15:27:20 -0800195{
Dave Hansencde70142016-02-12 13:01:55 -0800196 return __get_user_pages_unlocked(current, current->mm, start, nr_pages,
Lorenzo Stoakesc1641542016-10-13 01:20:13 +0100197 pages, gup_flags);
Andrea Arcangeli0fd71a52015-02-11 15:27:20 -0800198}
Ingo Molnarc12d2da2016-04-04 10:24:58 +0200199EXPORT_SYMBOL(get_user_pages_unlocked);
Andrea Arcangelif0818f42015-02-11 15:27:17 -0800200
Paul Mundtdfc2f912009-06-26 04:31:57 +0900201/**
202 * follow_pfn - look up PFN at a user virtual address
203 * @vma: memory mapping
204 * @address: user virtual address
205 * @pfn: location to store found PFN
206 *
207 * Only IO mappings and raw PFN mappings are allowed.
208 *
209 * Returns zero and the pfn at @pfn on success, -ve otherwise.
210 */
211int follow_pfn(struct vm_area_struct *vma, unsigned long address,
212 unsigned long *pfn)
213{
214 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
215 return -EINVAL;
216
217 *pfn = address >> PAGE_SHIFT;
218 return 0;
219}
220EXPORT_SYMBOL(follow_pfn);
221
Joonsoo Kimf1c40692013-04-29 15:07:37 -0700222LIST_HEAD(vmap_area_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800224void vfree(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225{
226 kfree(addr);
227}
Paul Mundtb5073172007-07-21 04:37:25 -0700228EXPORT_SYMBOL(vfree);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229
Al Virodd0fc662005-10-07 07:46:04 +0100230void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231{
232 /*
Robert P. J. Day85186092007-10-19 23:11:38 +0200233 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
234 * returns only a logical address.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 */
Nick Piggin84097512006-03-22 00:08:34 -0800236 return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237}
Paul Mundtb5073172007-07-21 04:37:25 -0700238EXPORT_SYMBOL(__vmalloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239
Michal Hockoa7c3e902017-05-08 15:57:09 -0700240void *__vmalloc_node_flags(unsigned long size, int node, gfp_t flags)
241{
242 return __vmalloc(size, flags, PAGE_KERNEL);
243}
244
Paul Mundtf905bc42008-02-04 22:29:59 -0800245void *vmalloc_user(unsigned long size)
246{
247 void *ret;
248
Michal Hocko19809c22017-05-08 15:57:44 -0700249 ret = __vmalloc(size, GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL);
Paul Mundtf905bc42008-02-04 22:29:59 -0800250 if (ret) {
251 struct vm_area_struct *vma;
252
253 down_write(&current->mm->mmap_sem);
254 vma = find_vma(current->mm, (unsigned long)ret);
255 if (vma)
256 vma->vm_flags |= VM_USERMAP;
257 up_write(&current->mm->mmap_sem);
258 }
259
260 return ret;
261}
262EXPORT_SYMBOL(vmalloc_user);
263
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800264struct page *vmalloc_to_page(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265{
266 return virt_to_page(addr);
267}
Paul Mundtb5073172007-07-21 04:37:25 -0700268EXPORT_SYMBOL(vmalloc_to_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800270unsigned long vmalloc_to_pfn(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271{
272 return page_to_pfn(virt_to_page(addr));
273}
Paul Mundtb5073172007-07-21 04:37:25 -0700274EXPORT_SYMBOL(vmalloc_to_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275
276long vread(char *buf, char *addr, unsigned long count)
277{
Chen Gang9bde9162013-07-03 15:02:36 -0700278 /* Don't allow overflow */
279 if ((unsigned long) buf + count < count)
280 count = -(unsigned long) buf;
281
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 memcpy(buf, addr, count);
283 return count;
284}
285
286long vwrite(char *buf, char *addr, unsigned long count)
287{
288 /* Don't allow overflow */
289 if ((unsigned long) addr + count < count)
290 count = -(unsigned long) addr;
291
292 memcpy(addr, buf, count);
Choi Gi-yongac714902014-04-07 15:37:36 -0700293 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294}
295
296/*
Masahiro Yamadae1c05062015-07-07 10:14:59 +0900297 * vmalloc - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298 *
299 * @size: allocation size
300 *
301 * Allocate enough pages to cover @size from the page level
Masahiro Yamadae1c05062015-07-07 10:14:59 +0900302 * allocator and map them into contiguous kernel virtual space.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +0200304 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305 * use __vmalloc() instead.
306 */
307void *vmalloc(unsigned long size)
308{
309 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
310}
Andrew Mortonf6138882006-02-28 16:59:18 -0800311EXPORT_SYMBOL(vmalloc);
312
Dave Younge1ca7782010-10-26 14:22:06 -0700313/*
Masahiro Yamadae1c05062015-07-07 10:14:59 +0900314 * vzalloc - allocate virtually contiguous memory with zero fill
Dave Younge1ca7782010-10-26 14:22:06 -0700315 *
316 * @size: allocation size
317 *
318 * Allocate enough pages to cover @size from the page level
Masahiro Yamadae1c05062015-07-07 10:14:59 +0900319 * allocator and map them into contiguous kernel virtual space.
Dave Younge1ca7782010-10-26 14:22:06 -0700320 * The memory allocated is set to zero.
321 *
322 * For tight control over page level allocator and protection flags
323 * use __vmalloc() instead.
324 */
325void *vzalloc(unsigned long size)
326{
327 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
328 PAGE_KERNEL);
329}
330EXPORT_SYMBOL(vzalloc);
331
332/**
333 * vmalloc_node - allocate memory on a specific node
334 * @size: allocation size
335 * @node: numa node
336 *
337 * Allocate enough pages to cover @size from the page level
338 * allocator and map them into contiguous kernel virtual space.
339 *
340 * For tight control over page level allocator and protection flags
341 * use __vmalloc() instead.
342 */
Andrew Mortonf6138882006-02-28 16:59:18 -0800343void *vmalloc_node(unsigned long size, int node)
344{
345 return vmalloc(size);
346}
Paul Mundt9a14f652010-12-24 11:50:34 +0900347EXPORT_SYMBOL(vmalloc_node);
Dave Younge1ca7782010-10-26 14:22:06 -0700348
349/**
350 * vzalloc_node - allocate memory on a specific node with zero fill
351 * @size: allocation size
352 * @node: numa node
353 *
354 * Allocate enough pages to cover @size from the page level
355 * allocator and map them into contiguous kernel virtual space.
356 * The memory allocated is set to zero.
357 *
358 * For tight control over page level allocator and protection flags
359 * use __vmalloc() instead.
360 */
361void *vzalloc_node(unsigned long size, int node)
362{
363 return vzalloc(size);
364}
365EXPORT_SYMBOL(vzalloc_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366
Paul Mundt1af446e2008-08-04 16:01:47 +0900367#ifndef PAGE_KERNEL_EXEC
368# define PAGE_KERNEL_EXEC PAGE_KERNEL
369#endif
370
371/**
372 * vmalloc_exec - allocate virtually contiguous, executable memory
373 * @size: allocation size
374 *
375 * Kernel-internal function to allocate enough pages to cover @size
376 * the page level allocator and map them into contiguous and
377 * executable kernel virtual space.
378 *
379 * For tight control over page level allocator and protection flags
380 * use __vmalloc() instead.
381 */
382
383void *vmalloc_exec(unsigned long size)
384{
385 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
386}
387
Paul Mundtb5073172007-07-21 04:37:25 -0700388/**
389 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 * @size: allocation size
391 *
392 * Allocate enough 32bit PA addressable pages to cover @size from the
Masahiro Yamadae1c05062015-07-07 10:14:59 +0900393 * page level allocator and map them into contiguous kernel virtual space.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 */
395void *vmalloc_32(unsigned long size)
396{
397 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
398}
Paul Mundtb5073172007-07-21 04:37:25 -0700399EXPORT_SYMBOL(vmalloc_32);
400
401/**
402 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
403 * @size: allocation size
404 *
405 * The resulting memory area is 32bit addressable and zeroed so it can be
406 * mapped to userspace without leaking data.
Paul Mundtf905bc42008-02-04 22:29:59 -0800407 *
408 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
409 * remap_vmalloc_range() are permissible.
Paul Mundtb5073172007-07-21 04:37:25 -0700410 */
411void *vmalloc_32_user(unsigned long size)
412{
Paul Mundtf905bc42008-02-04 22:29:59 -0800413 /*
414 * We'll have to sort out the ZONE_DMA bits for 64-bit,
415 * but for now this can simply use vmalloc_user() directly.
416 */
417 return vmalloc_user(size);
Paul Mundtb5073172007-07-21 04:37:25 -0700418}
419EXPORT_SYMBOL(vmalloc_32_user);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420
421void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
422{
423 BUG();
424 return NULL;
425}
Paul Mundtb5073172007-07-21 04:37:25 -0700426EXPORT_SYMBOL(vmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800428void vunmap(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429{
430 BUG();
431}
Paul Mundtb5073172007-07-21 04:37:25 -0700432EXPORT_SYMBOL(vunmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433
Paul Mundteb6434d2009-01-21 17:45:47 +0900434void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
435{
436 BUG();
437 return NULL;
438}
439EXPORT_SYMBOL(vm_map_ram);
440
441void vm_unmap_ram(const void *mem, unsigned int count)
442{
443 BUG();
444}
445EXPORT_SYMBOL(vm_unmap_ram);
446
447void vm_unmap_aliases(void)
448{
449}
450EXPORT_SYMBOL_GPL(vm_unmap_aliases);
451
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452/*
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -0700453 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
454 * have one.
455 */
Gideon Israel Dsouza3b321232014-04-07 15:37:26 -0700456void __weak vmalloc_sync_all(void)
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -0700457{
458}
459
Paul Mundt29c185e2010-12-24 12:08:30 +0900460/**
461 * alloc_vm_area - allocate a range of kernel address space
462 * @size: size of the area
463 *
464 * Returns: NULL on failure, vm_struct on success
465 *
466 * This function reserves a range of kernel address space, and
467 * allocates pagetables to map that range. No actual mappings
468 * are created. If the kernel address space is not shared
469 * between processes, it syncs the pagetable across all
470 * processes.
471 */
David Vrabelcd129092011-09-29 16:53:32 +0100472struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
Paul Mundt29c185e2010-12-24 12:08:30 +0900473{
474 BUG();
475 return NULL;
476}
477EXPORT_SYMBOL_GPL(alloc_vm_area);
478
479void free_vm_area(struct vm_struct *area)
480{
481 BUG();
482}
483EXPORT_SYMBOL_GPL(free_vm_area);
484
Paul Mundtb5073172007-07-21 04:37:25 -0700485int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
486 struct page *page)
487{
488 return -EINVAL;
489}
490EXPORT_SYMBOL(vm_insert_page);
491
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -0700492/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493 * sys_brk() for the most part doesn't need the global kernel
494 * lock, except when an application is doing something nasty
495 * like trying to un-brk an area that has already been mapped
496 * to a regular file. in this case, the unmapping will need
497 * to invoke file system routines that need the global lock.
498 */
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100499SYSCALL_DEFINE1(brk, unsigned long, brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500{
501 struct mm_struct *mm = current->mm;
502
503 if (brk < mm->start_brk || brk > mm->context.end_brk)
504 return mm->brk;
505
506 if (mm->brk == brk)
507 return mm->brk;
508
509 /*
510 * Always allow shrinking brk
511 */
512 if (brk <= mm->brk) {
513 mm->brk = brk;
514 return brk;
515 }
516
517 /*
518 * Ok, looks good - let it rip.
519 */
Mike Frysingercfe79c02010-01-06 17:23:23 +0000520 flush_icache_range(mm->brk, brk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 return mm->brk = brk;
522}
523
David Howells8feae132009-01-08 12:04:47 +0000524/*
seokhoon.yoon3edf41d2017-02-24 14:56:44 -0800525 * initialise the percpu counter for VM and region record slabs
David Howells8feae132009-01-08 12:04:47 +0000526 */
527void __init mmap_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528{
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -0700529 int ret;
530
Tejun Heo908c7f12014-09-08 09:51:29 +0900531 ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -0700532 VM_BUG_ON(ret);
Vladimir Davydov5d097052016-01-14 15:18:21 -0800533 vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC|SLAB_ACCOUNT);
David Howells8feae132009-01-08 12:04:47 +0000534}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535
David Howells8feae132009-01-08 12:04:47 +0000536/*
537 * validate the region tree
538 * - the caller must hold the region lock
539 */
540#ifdef CONFIG_DEBUG_NOMMU_REGIONS
541static noinline void validate_nommu_regions(void)
542{
543 struct vm_region *region, *last;
544 struct rb_node *p, *lastp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545
David Howells8feae132009-01-08 12:04:47 +0000546 lastp = rb_first(&nommu_region_tree);
547 if (!lastp)
548 return;
549
550 last = rb_entry(lastp, struct vm_region, vm_rb);
Geliang Tangc9427bc2015-11-05 18:48:38 -0800551 BUG_ON(last->vm_end <= last->vm_start);
552 BUG_ON(last->vm_top < last->vm_end);
David Howells8feae132009-01-08 12:04:47 +0000553
554 while ((p = rb_next(lastp))) {
555 region = rb_entry(p, struct vm_region, vm_rb);
556 last = rb_entry(lastp, struct vm_region, vm_rb);
557
Geliang Tangc9427bc2015-11-05 18:48:38 -0800558 BUG_ON(region->vm_end <= region->vm_start);
559 BUG_ON(region->vm_top < region->vm_end);
560 BUG_ON(region->vm_start < last->vm_top);
David Howells8feae132009-01-08 12:04:47 +0000561
562 lastp = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 }
564}
David Howells8feae132009-01-08 12:04:47 +0000565#else
David Howells33e5d7692009-04-02 16:56:32 -0700566static void validate_nommu_regions(void)
567{
568}
David Howells8feae132009-01-08 12:04:47 +0000569#endif
570
571/*
572 * add a region into the global tree
573 */
574static void add_nommu_region(struct vm_region *region)
575{
576 struct vm_region *pregion;
577 struct rb_node **p, *parent;
578
579 validate_nommu_regions();
580
David Howells8feae132009-01-08 12:04:47 +0000581 parent = NULL;
582 p = &nommu_region_tree.rb_node;
583 while (*p) {
584 parent = *p;
585 pregion = rb_entry(parent, struct vm_region, vm_rb);
586 if (region->vm_start < pregion->vm_start)
587 p = &(*p)->rb_left;
588 else if (region->vm_start > pregion->vm_start)
589 p = &(*p)->rb_right;
590 else if (pregion == region)
591 return;
592 else
593 BUG();
594 }
595
596 rb_link_node(&region->vm_rb, parent, p);
597 rb_insert_color(&region->vm_rb, &nommu_region_tree);
598
599 validate_nommu_regions();
600}
601
602/*
603 * delete a region from the global tree
604 */
605static void delete_nommu_region(struct vm_region *region)
606{
607 BUG_ON(!nommu_region_tree.rb_node);
608
609 validate_nommu_regions();
610 rb_erase(&region->vm_rb, &nommu_region_tree);
611 validate_nommu_regions();
612}
613
614/*
615 * free a contiguous series of pages
616 */
617static void free_page_series(unsigned long from, unsigned long to)
618{
619 for (; from < to; from += PAGE_SIZE) {
620 struct page *page = virt_to_page(from);
621
David Howells33e5d7692009-04-02 16:56:32 -0700622 atomic_long_dec(&mmap_pages_allocated);
David Howells8feae132009-01-08 12:04:47 +0000623 put_page(page);
624 }
625}
626
627/*
628 * release a reference to a region
David Howells33e5d7692009-04-02 16:56:32 -0700629 * - the caller must hold the region semaphore for writing, which this releases
Paul Mundtdd8632a2009-01-08 12:04:47 +0000630 * - the region may not have been added to the tree yet, in which case vm_top
David Howells8feae132009-01-08 12:04:47 +0000631 * will equal vm_start
632 */
633static void __put_nommu_region(struct vm_region *region)
634 __releases(nommu_region_sem)
635{
David Howells8feae132009-01-08 12:04:47 +0000636 BUG_ON(!nommu_region_tree.rb_node);
637
David Howells1e2ae592010-01-15 17:01:33 -0800638 if (--region->vm_usage == 0) {
Paul Mundtdd8632a2009-01-08 12:04:47 +0000639 if (region->vm_top > region->vm_start)
David Howells8feae132009-01-08 12:04:47 +0000640 delete_nommu_region(region);
641 up_write(&nommu_region_sem);
642
643 if (region->vm_file)
644 fput(region->vm_file);
645
646 /* IO memory and memory shared directly out of the pagecache
647 * from ramfs/tmpfs mustn't be released here */
Leon Romanovsky22cc8772015-06-24 16:57:47 -0700648 if (region->vm_flags & VM_MAPPED_COPY)
Paul Mundtdd8632a2009-01-08 12:04:47 +0000649 free_page_series(region->vm_start, region->vm_top);
David Howells8feae132009-01-08 12:04:47 +0000650 kmem_cache_free(vm_region_jar, region);
651 } else {
652 up_write(&nommu_region_sem);
653 }
654}
655
656/*
657 * release a reference to a region
658 */
659static void put_nommu_region(struct vm_region *region)
660{
661 down_write(&nommu_region_sem);
662 __put_nommu_region(region);
663}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664
David Howells30340972006-09-27 01:50:20 -0700665/*
Bernd Schmidteb8cdec2009-09-21 17:03:57 -0700666 * update protection on a vma
667 */
668static void protect_vma(struct vm_area_struct *vma, unsigned long flags)
669{
670#ifdef CONFIG_MPU
671 struct mm_struct *mm = vma->vm_mm;
672 long start = vma->vm_start & PAGE_MASK;
673 while (start < vma->vm_end) {
674 protect_page(mm, start, flags);
675 start += PAGE_SIZE;
676 }
677 update_protections(mm);
678#endif
679}
680
681/*
David Howells30340972006-09-27 01:50:20 -0700682 * add a VMA into a process's mm_struct in the appropriate place in the list
David Howells8feae132009-01-08 12:04:47 +0000683 * and tree and add to the address space's page tree also if not an anonymous
684 * page
David Howells30340972006-09-27 01:50:20 -0700685 * - should be called with mm->mmap_sem held writelocked
686 */
David Howells8feae132009-01-08 12:04:47 +0000687static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
David Howells30340972006-09-27 01:50:20 -0700688{
Namhyung Kim6038def2011-05-24 17:11:22 -0700689 struct vm_area_struct *pvma, *prev;
David Howells8feae132009-01-08 12:04:47 +0000690 struct address_space *mapping;
Namhyung Kim6038def2011-05-24 17:11:22 -0700691 struct rb_node **p, *parent, *rb_prev;
David Howells30340972006-09-27 01:50:20 -0700692
David Howells8feae132009-01-08 12:04:47 +0000693 BUG_ON(!vma->vm_region);
694
695 mm->map_count++;
696 vma->vm_mm = mm;
697
Bernd Schmidteb8cdec2009-09-21 17:03:57 -0700698 protect_vma(vma, vma->vm_flags);
699
David Howells8feae132009-01-08 12:04:47 +0000700 /* add the VMA to the mapping */
701 if (vma->vm_file) {
702 mapping = vma->vm_file->f_mapping;
703
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800704 i_mmap_lock_write(mapping);
David Howells8feae132009-01-08 12:04:47 +0000705 flush_dcache_mmap_lock(mapping);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700706 vma_interval_tree_insert(vma, &mapping->i_mmap);
David Howells8feae132009-01-08 12:04:47 +0000707 flush_dcache_mmap_unlock(mapping);
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800708 i_mmap_unlock_write(mapping);
David Howells8feae132009-01-08 12:04:47 +0000709 }
710
711 /* add the VMA to the tree */
Namhyung Kim6038def2011-05-24 17:11:22 -0700712 parent = rb_prev = NULL;
David Howells8feae132009-01-08 12:04:47 +0000713 p = &mm->mm_rb.rb_node;
714 while (*p) {
715 parent = *p;
716 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
717
718 /* sort by: start addr, end addr, VMA struct addr in that order
719 * (the latter is necessary as we may get identical VMAs) */
720 if (vma->vm_start < pvma->vm_start)
721 p = &(*p)->rb_left;
Namhyung Kim6038def2011-05-24 17:11:22 -0700722 else if (vma->vm_start > pvma->vm_start) {
723 rb_prev = parent;
David Howells8feae132009-01-08 12:04:47 +0000724 p = &(*p)->rb_right;
Namhyung Kim6038def2011-05-24 17:11:22 -0700725 } else if (vma->vm_end < pvma->vm_end)
David Howells8feae132009-01-08 12:04:47 +0000726 p = &(*p)->rb_left;
Namhyung Kim6038def2011-05-24 17:11:22 -0700727 else if (vma->vm_end > pvma->vm_end) {
728 rb_prev = parent;
David Howells8feae132009-01-08 12:04:47 +0000729 p = &(*p)->rb_right;
Namhyung Kim6038def2011-05-24 17:11:22 -0700730 } else if (vma < pvma)
David Howells8feae132009-01-08 12:04:47 +0000731 p = &(*p)->rb_left;
Namhyung Kim6038def2011-05-24 17:11:22 -0700732 else if (vma > pvma) {
733 rb_prev = parent;
David Howells8feae132009-01-08 12:04:47 +0000734 p = &(*p)->rb_right;
Namhyung Kim6038def2011-05-24 17:11:22 -0700735 } else
David Howells8feae132009-01-08 12:04:47 +0000736 BUG();
737 }
738
739 rb_link_node(&vma->vm_rb, parent, p);
740 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
741
742 /* add VMA to the VMA list also */
Namhyung Kim6038def2011-05-24 17:11:22 -0700743 prev = NULL;
744 if (rb_prev)
745 prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
David Howells30340972006-09-27 01:50:20 -0700746
Namhyung Kim6038def2011-05-24 17:11:22 -0700747 __vma_link_list(mm, vma, prev, parent);
David Howells8feae132009-01-08 12:04:47 +0000748}
749
750/*
751 * delete a VMA from its owning mm_struct and address space
752 */
753static void delete_vma_from_mm(struct vm_area_struct *vma)
754{
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700755 int i;
David Howells8feae132009-01-08 12:04:47 +0000756 struct address_space *mapping;
757 struct mm_struct *mm = vma->vm_mm;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700758 struct task_struct *curr = current;
David Howells8feae132009-01-08 12:04:47 +0000759
Bernd Schmidteb8cdec2009-09-21 17:03:57 -0700760 protect_vma(vma, 0);
761
David Howells8feae132009-01-08 12:04:47 +0000762 mm->map_count--;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700763 for (i = 0; i < VMACACHE_SIZE; i++) {
764 /* if the vma is cached, invalidate the entire cache */
Ingo Molnar314ff782017-02-03 11:03:31 +0100765 if (curr->vmacache.vmas[i] == vma) {
Steven Miaoe020d5b2014-06-23 13:22:02 -0700766 vmacache_invalidate(mm);
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700767 break;
768 }
769 }
David Howells8feae132009-01-08 12:04:47 +0000770
771 /* remove the VMA from the mapping */
772 if (vma->vm_file) {
773 mapping = vma->vm_file->f_mapping;
774
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800775 i_mmap_lock_write(mapping);
David Howells8feae132009-01-08 12:04:47 +0000776 flush_dcache_mmap_lock(mapping);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700777 vma_interval_tree_remove(vma, &mapping->i_mmap);
David Howells8feae132009-01-08 12:04:47 +0000778 flush_dcache_mmap_unlock(mapping);
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800779 i_mmap_unlock_write(mapping);
David Howells8feae132009-01-08 12:04:47 +0000780 }
781
782 /* remove from the MM's tree and list */
783 rb_erase(&vma->vm_rb, &mm->mm_rb);
Namhyung Kimb951bf22011-05-24 17:11:23 -0700784
785 if (vma->vm_prev)
786 vma->vm_prev->vm_next = vma->vm_next;
787 else
788 mm->mmap = vma->vm_next;
789
790 if (vma->vm_next)
791 vma->vm_next->vm_prev = vma->vm_prev;
David Howells8feae132009-01-08 12:04:47 +0000792}
793
794/*
795 * destroy a VMA record
796 */
797static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
798{
David Howells8feae132009-01-08 12:04:47 +0000799 if (vma->vm_ops && vma->vm_ops->close)
800 vma->vm_ops->close(vma);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -0700801 if (vma->vm_file)
David Howells8feae132009-01-08 12:04:47 +0000802 fput(vma->vm_file);
David Howells8feae132009-01-08 12:04:47 +0000803 put_nommu_region(vma->vm_region);
804 kmem_cache_free(vm_area_cachep, vma);
David Howells30340972006-09-27 01:50:20 -0700805}
806
807/*
808 * look up the first VMA in which addr resides, NULL if none
809 * - should be called with mm->mmap_sem at least held readlocked
810 */
811struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
812{
David Howells8feae132009-01-08 12:04:47 +0000813 struct vm_area_struct *vma;
David Howells30340972006-09-27 01:50:20 -0700814
David Howells8feae132009-01-08 12:04:47 +0000815 /* check the cache first */
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700816 vma = vmacache_find(mm, addr);
817 if (likely(vma))
David Howells8feae132009-01-08 12:04:47 +0000818 return vma;
819
Namhyung Kime922c4c2011-05-24 17:11:24 -0700820 /* trawl the list (there may be multiple mappings in which addr
David Howells8feae132009-01-08 12:04:47 +0000821 * resides) */
Namhyung Kime922c4c2011-05-24 17:11:24 -0700822 for (vma = mm->mmap; vma; vma = vma->vm_next) {
David Howells8feae132009-01-08 12:04:47 +0000823 if (vma->vm_start > addr)
824 return NULL;
825 if (vma->vm_end > addr) {
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700826 vmacache_update(addr, vma);
David Howells8feae132009-01-08 12:04:47 +0000827 return vma;
828 }
David Howells30340972006-09-27 01:50:20 -0700829 }
830
David Howells30340972006-09-27 01:50:20 -0700831 return NULL;
832}
833EXPORT_SYMBOL(find_vma);
834
835/*
David Howells930e6522006-09-27 01:50:22 -0700836 * find a VMA
837 * - we don't extend stack VMAs under NOMMU conditions
838 */
839struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
840{
David Howells7561e8c2010-03-25 16:48:38 +0000841 return find_vma(mm, addr);
David Howells930e6522006-09-27 01:50:22 -0700842}
843
David Howells8feae132009-01-08 12:04:47 +0000844/*
845 * expand a stack to a given address
846 * - not supported under NOMMU conditions
847 */
Greg Ungerer57c8f632007-07-15 23:38:28 -0700848int expand_stack(struct vm_area_struct *vma, unsigned long address)
849{
850 return -ENOMEM;
851}
852
David Howells930e6522006-09-27 01:50:22 -0700853/*
David Howells6fa5f802006-09-27 01:50:21 -0700854 * look up the first VMA exactly that exactly matches addr
855 * - should be called with mm->mmap_sem at least held readlocked
856 */
David Howells8feae132009-01-08 12:04:47 +0000857static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
858 unsigned long addr,
859 unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860{
861 struct vm_area_struct *vma;
David Howells8feae132009-01-08 12:04:47 +0000862 unsigned long end = addr + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863
David Howells8feae132009-01-08 12:04:47 +0000864 /* check the cache first */
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700865 vma = vmacache_find_exact(mm, addr, end);
866 if (vma)
David Howells8feae132009-01-08 12:04:47 +0000867 return vma;
868
Namhyung Kime922c4c2011-05-24 17:11:24 -0700869 /* trawl the list (there may be multiple mappings in which addr
David Howells8feae132009-01-08 12:04:47 +0000870 * resides) */
Namhyung Kime922c4c2011-05-24 17:11:24 -0700871 for (vma = mm->mmap; vma; vma = vma->vm_next) {
David Howells8feae132009-01-08 12:04:47 +0000872 if (vma->vm_start < addr)
873 continue;
874 if (vma->vm_start > addr)
875 return NULL;
876 if (vma->vm_end == end) {
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700877 vmacache_update(addr, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 return vma;
David Howells8feae132009-01-08 12:04:47 +0000879 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 }
881
882 return NULL;
883}
884
David Howells30340972006-09-27 01:50:20 -0700885/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 * determine whether a mapping should be permitted and, if so, what sort of
887 * mapping we're capable of supporting
888 */
889static int validate_mmap_request(struct file *file,
890 unsigned long addr,
891 unsigned long len,
892 unsigned long prot,
893 unsigned long flags,
894 unsigned long pgoff,
895 unsigned long *_capabilities)
896{
David Howells8feae132009-01-08 12:04:47 +0000897 unsigned long capabilities, rlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 int ret;
899
900 /* do the simple checks first */
Leon Romanovsky22cc8772015-06-24 16:57:47 -0700901 if (flags & MAP_FIXED)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903
904 if ((flags & MAP_TYPE) != MAP_PRIVATE &&
905 (flags & MAP_TYPE) != MAP_SHARED)
906 return -EINVAL;
907
Mike Frysingerf81cff0d2006-12-06 12:02:59 +1000908 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 return -EINVAL;
910
Mike Frysingerf81cff0d2006-12-06 12:02:59 +1000911 /* Careful about overflows.. */
David Howells8feae132009-01-08 12:04:47 +0000912 rlen = PAGE_ALIGN(len);
913 if (!rlen || rlen > TASK_SIZE)
Mike Frysingerf81cff0d2006-12-06 12:02:59 +1000914 return -ENOMEM;
915
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 /* offset overflow? */
David Howells8feae132009-01-08 12:04:47 +0000917 if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
Mike Frysingerf81cff0d2006-12-06 12:02:59 +1000918 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919
920 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 /* files must support mmap */
Al Viro72c2d532013-09-22 16:27:52 -0400922 if (!file->f_op->mmap)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 return -ENODEV;
924
925 /* work out if what we've got could possibly be shared
926 * - we support chardevs that provide their own "memory"
927 * - we support files/blockdevs that are memory backed
928 */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100929 if (file->f_op->mmap_capabilities) {
930 capabilities = file->f_op->mmap_capabilities(file);
931 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 /* no explicit capabilities set, so assume some
933 * defaults */
Al Viro496ad9a2013-01-23 17:07:38 -0500934 switch (file_inode(file)->i_mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 case S_IFREG:
936 case S_IFBLK:
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100937 capabilities = NOMMU_MAP_COPY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938 break;
939
940 case S_IFCHR:
941 capabilities =
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100942 NOMMU_MAP_DIRECT |
943 NOMMU_MAP_READ |
944 NOMMU_MAP_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 break;
946
947 default:
948 return -EINVAL;
949 }
950 }
951
952 /* eliminate any capabilities that we can't support on this
953 * device */
954 if (!file->f_op->get_unmapped_area)
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100955 capabilities &= ~NOMMU_MAP_DIRECT;
Al Viro6e242a12015-03-31 12:35:13 -0400956 if (!(file->f_mode & FMODE_CAN_READ))
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100957 capabilities &= ~NOMMU_MAP_COPY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958
Graff Yang28d7a6a2009-08-18 14:11:17 -0700959 /* The file shall have been opened with read permission. */
960 if (!(file->f_mode & FMODE_READ))
961 return -EACCES;
962
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 if (flags & MAP_SHARED) {
964 /* do checks for writing, appending and locking */
965 if ((prot & PROT_WRITE) &&
966 !(file->f_mode & FMODE_WRITE))
967 return -EACCES;
968
Al Viro496ad9a2013-01-23 17:07:38 -0500969 if (IS_APPEND(file_inode(file)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 (file->f_mode & FMODE_WRITE))
971 return -EACCES;
972
Jeff Laytond7a06982014-03-10 09:54:15 -0400973 if (locks_verify_locked(file))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 return -EAGAIN;
975
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100976 if (!(capabilities & NOMMU_MAP_DIRECT))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 return -ENODEV;
978
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 /* we mustn't privatise shared mappings */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100980 capabilities &= ~NOMMU_MAP_COPY;
Choi Gi-yongac714902014-04-07 15:37:36 -0700981 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 /* we're going to read the file into private memory we
983 * allocate */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100984 if (!(capabilities & NOMMU_MAP_COPY))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 return -ENODEV;
986
987 /* we don't permit a private writable mapping to be
988 * shared with the backing device */
989 if (prot & PROT_WRITE)
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100990 capabilities &= ~NOMMU_MAP_DIRECT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 }
992
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100993 if (capabilities & NOMMU_MAP_DIRECT) {
994 if (((prot & PROT_READ) && !(capabilities & NOMMU_MAP_READ)) ||
995 ((prot & PROT_WRITE) && !(capabilities & NOMMU_MAP_WRITE)) ||
996 ((prot & PROT_EXEC) && !(capabilities & NOMMU_MAP_EXEC))
Bernd Schmidt3c7b2042010-05-25 23:43:00 -0700997 ) {
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100998 capabilities &= ~NOMMU_MAP_DIRECT;
Bernd Schmidt3c7b2042010-05-25 23:43:00 -0700999 if (flags & MAP_SHARED) {
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001000 pr_warn("MAP_SHARED not completely supported on !MMU\n");
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001001 return -EINVAL;
1002 }
1003 }
1004 }
1005
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 /* handle executable mappings and implied executable
1007 * mappings */
Eric W. Biederman90f85722015-06-29 14:42:03 -05001008 if (path_noexec(&file->f_path)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 if (prot & PROT_EXEC)
1010 return -EPERM;
Choi Gi-yongac714902014-04-07 15:37:36 -07001011 } else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 /* handle implication of PROT_EXEC by PROT_READ */
1013 if (current->personality & READ_IMPLIES_EXEC) {
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001014 if (capabilities & NOMMU_MAP_EXEC)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 prot |= PROT_EXEC;
1016 }
Choi Gi-yongac714902014-04-07 15:37:36 -07001017 } else if ((prot & PROT_READ) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 (prot & PROT_EXEC) &&
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001019 !(capabilities & NOMMU_MAP_EXEC)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 ) {
1021 /* backing file is not executable, try to copy */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001022 capabilities &= ~NOMMU_MAP_DIRECT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 }
Choi Gi-yongac714902014-04-07 15:37:36 -07001024 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 /* anonymous mappings are always memory backed and can be
1026 * privately mapped
1027 */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001028 capabilities = NOMMU_MAP_COPY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029
1030 /* handle PROT_EXEC implication by PROT_READ */
1031 if ((prot & PROT_READ) &&
1032 (current->personality & READ_IMPLIES_EXEC))
1033 prot |= PROT_EXEC;
1034 }
1035
1036 /* allow the security API to have its say */
Al Viroe5467852012-05-30 13:30:51 -04001037 ret = security_mmap_addr(addr);
1038 if (ret < 0)
1039 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040
1041 /* looks okay */
1042 *_capabilities = capabilities;
1043 return 0;
1044}
1045
1046/*
1047 * we've determined that we can make the mapping, now translate what we
1048 * now know into VMA flags
1049 */
1050static unsigned long determine_vm_flags(struct file *file,
1051 unsigned long prot,
1052 unsigned long flags,
1053 unsigned long capabilities)
1054{
1055 unsigned long vm_flags;
1056
Dave Hansene6bfb702016-02-12 13:02:31 -08001057 vm_flags = calc_vm_prot_bits(prot, 0) | calc_vm_flag_bits(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 /* vm_flags |= mm->def_flags; */
1059
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001060 if (!(capabilities & NOMMU_MAP_DIRECT)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 /* attempt to share read-only copies of mapped file chunks */
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001062 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 if (file && !(prot & PROT_WRITE))
1064 vm_flags |= VM_MAYSHARE;
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001065 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 /* overlay a shareable mapping on the backing device or inode
1067 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1068 * romfs/cramfs */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001069 vm_flags |= VM_MAYSHARE | (capabilities & NOMMU_VMFLAGS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 if (flags & MAP_SHARED)
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001071 vm_flags |= VM_SHARED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 }
1073
1074 /* refuse to let anyone share private mappings with this process if
1075 * it's being traced - otherwise breakpoints set in it may interfere
1076 * with another untraced process
1077 */
Tejun Heoa288eec2011-06-17 16:50:37 +02001078 if ((flags & MAP_PRIVATE) && current->ptrace)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 vm_flags &= ~VM_MAYSHARE;
1080
1081 return vm_flags;
1082}
1083
1084/*
David Howells8feae132009-01-08 12:04:47 +00001085 * set up a shared mapping on a file (the driver or filesystem provides and
1086 * pins the storage)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 */
David Howells8feae132009-01-08 12:04:47 +00001088static int do_mmap_shared_file(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089{
1090 int ret;
1091
Miklos Szeredif74ac012017-02-20 16:51:23 +01001092 ret = call_mmap(vma->vm_file, vma);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001093 if (ret == 0) {
1094 vma->vm_region->vm_top = vma->vm_region->vm_end;
David Howells645d83c2009-09-24 15:13:10 +01001095 return 0;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001096 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 if (ret != -ENOSYS)
1098 return ret;
1099
David Howells3fa30462010-03-23 13:35:21 -07001100 /* getting -ENOSYS indicates that direct mmap isn't possible (as
1101 * opposed to tried but failed) so we can only give a suitable error as
1102 * it's not possible to make a private copy if MAP_SHARED was given */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 return -ENODEV;
1104}
1105
1106/*
1107 * set up a private mapping or an anonymous shared mapping
1108 */
David Howells8feae132009-01-08 12:04:47 +00001109static int do_mmap_private(struct vm_area_struct *vma,
1110 struct vm_region *region,
David Howells645d83c2009-09-24 15:13:10 +01001111 unsigned long len,
1112 unsigned long capabilities)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113{
Joonsoo Kimdbc83582014-12-12 16:55:55 -08001114 unsigned long total, point;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 void *base;
David Howells8feae132009-01-08 12:04:47 +00001116 int ret, order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117
1118 /* invoke the file's mapping function so that it can keep track of
1119 * shared mappings on devices or memory
1120 * - VM_MAYSHARE will be set if it may attempt to share
1121 */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001122 if (capabilities & NOMMU_MAP_DIRECT) {
Miklos Szeredif74ac012017-02-20 16:51:23 +01001123 ret = call_mmap(vma->vm_file, vma);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001124 if (ret == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 /* shouldn't return success if we're not sharing */
Paul Mundtdd8632a2009-01-08 12:04:47 +00001126 BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1127 vma->vm_region->vm_top = vma->vm_region->vm_end;
David Howells645d83c2009-09-24 15:13:10 +01001128 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 }
Paul Mundtdd8632a2009-01-08 12:04:47 +00001130 if (ret != -ENOSYS)
1131 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132
1133 /* getting an ENOSYS error indicates that direct mmap isn't
1134 * possible (as opposed to tried but failed) so we'll try to
1135 * make a private copy of the data and map that instead */
1136 }
1137
David Howells8feae132009-01-08 12:04:47 +00001138
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139 /* allocate some memory to hold the mapping
1140 * - note that this may not return a page-aligned address if the object
1141 * we're allocating is smaller than a page
1142 */
Bob Liuf67d9b12011-05-24 17:12:56 -07001143 order = get_order(len);
David Howells8feae132009-01-08 12:04:47 +00001144 total = 1 << order;
Bob Liuf67d9b12011-05-24 17:12:56 -07001145 point = len >> PAGE_SHIFT;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001146
Joonsoo Kimdbc83582014-12-12 16:55:55 -08001147 /* we don't want to allocate a power-of-2 sized page set */
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001148 if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages)
Joonsoo Kimdbc83582014-12-12 16:55:55 -08001149 total = point;
David Howells8feae132009-01-08 12:04:47 +00001150
Joonsoo Kimda616532015-02-27 15:51:43 -08001151 base = alloc_pages_exact(total << PAGE_SHIFT, GFP_KERNEL);
Joonsoo Kimdbc83582014-12-12 16:55:55 -08001152 if (!base)
1153 goto enomem;
David Howells8feae132009-01-08 12:04:47 +00001154
Joonsoo Kimdbc83582014-12-12 16:55:55 -08001155 atomic_long_add(total, &mmap_pages_allocated);
1156
David Howells8feae132009-01-08 12:04:47 +00001157 region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1158 region->vm_start = (unsigned long) base;
Bob Liuf67d9b12011-05-24 17:12:56 -07001159 region->vm_end = region->vm_start + len;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001160 region->vm_top = region->vm_start + (total << PAGE_SHIFT);
David Howells8feae132009-01-08 12:04:47 +00001161
1162 vma->vm_start = region->vm_start;
1163 vma->vm_end = region->vm_start + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164
1165 if (vma->vm_file) {
1166 /* read the contents of a file into the copy */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167 loff_t fpos;
1168
1169 fpos = vma->vm_pgoff;
1170 fpos <<= PAGE_SHIFT;
1171
Christoph Hellwigb4bf8022017-09-01 17:39:17 +02001172 ret = kernel_read(vma->vm_file, base, len, &fpos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 if (ret < 0)
1174 goto error_free;
1175
1176 /* clear the last little bit */
Bob Liuf67d9b12011-05-24 17:12:56 -07001177 if (ret < len)
1178 memset(base + ret, 0, len - ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180 }
1181
1182 return 0;
1183
1184error_free:
Namhyung Kim7223bb42011-05-24 17:11:26 -07001185 free_page_series(region->vm_start, region->vm_top);
David Howells8feae132009-01-08 12:04:47 +00001186 region->vm_start = vma->vm_start = 0;
1187 region->vm_end = vma->vm_end = 0;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001188 region->vm_top = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189 return ret;
1190
1191enomem:
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -07001192 pr_err("Allocation of length %lu from process %d (%s) failed\n",
Greg Ungerer05ae6fa2009-01-13 17:30:22 +10001193 len, current->pid, current->comm);
Michal Hocko9af744d2017-02-22 15:46:16 -08001194 show_free_areas(0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 return -ENOMEM;
1196}
1197
1198/*
1199 * handle mapping creation for uClinux
1200 */
Oleg Nesterov1fcfd8d2015-09-09 15:39:29 -07001201unsigned long do_mmap(struct file *file,
1202 unsigned long addr,
1203 unsigned long len,
1204 unsigned long prot,
1205 unsigned long flags,
1206 vm_flags_t vm_flags,
1207 unsigned long pgoff,
Mike Rapoport897ab3e2017-02-24 14:58:22 -08001208 unsigned long *populate,
1209 struct list_head *uf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210{
David Howells8feae132009-01-08 12:04:47 +00001211 struct vm_area_struct *vma;
1212 struct vm_region *region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 struct rb_node *rb;
Oleg Nesterov1fcfd8d2015-09-09 15:39:29 -07001214 unsigned long capabilities, result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 int ret;
1216
Michel Lespinasse41badc12013-02-22 16:32:47 -08001217 *populate = 0;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001218
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 /* decide whether we should attempt the mapping, and if so what sort of
1220 * mapping */
1221 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1222 &capabilities);
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001223 if (ret < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 return ret;
1225
David Howells06aab5a2009-09-24 12:33:48 +01001226 /* we ignore the address hint */
1227 addr = 0;
Bob Liuf67d9b12011-05-24 17:12:56 -07001228 len = PAGE_ALIGN(len);
David Howells06aab5a2009-09-24 12:33:48 +01001229
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 /* we've determined that we can make the mapping, now translate what we
1231 * now know into VMA flags */
Oleg Nesterov1fcfd8d2015-09-09 15:39:29 -07001232 vm_flags |= determine_vm_flags(file, prot, flags, capabilities);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233
David Howells8feae132009-01-08 12:04:47 +00001234 /* we're going to need to record the mapping */
1235 region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1236 if (!region)
1237 goto error_getting_region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238
David Howells8feae132009-01-08 12:04:47 +00001239 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1240 if (!vma)
1241 goto error_getting_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242
David Howells1e2ae592010-01-15 17:01:33 -08001243 region->vm_usage = 1;
David Howells8feae132009-01-08 12:04:47 +00001244 region->vm_flags = vm_flags;
1245 region->vm_pgoff = pgoff;
1246
Rik van Riel5beb4932010-03-05 13:42:07 -08001247 INIT_LIST_HEAD(&vma->anon_vma_chain);
David Howells8feae132009-01-08 12:04:47 +00001248 vma->vm_flags = vm_flags;
1249 vma->vm_pgoff = pgoff;
1250
1251 if (file) {
Al Virocb0942b2012-08-27 14:48:26 -04001252 region->vm_file = get_file(file);
1253 vma->vm_file = get_file(file);
David Howells8feae132009-01-08 12:04:47 +00001254 }
1255
1256 down_write(&nommu_region_sem);
1257
1258 /* if we want to share, we need to check for regions created by other
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 * mmap() calls that overlap with our proposed mapping
David Howells8feae132009-01-08 12:04:47 +00001260 * - we can only share with a superset match on most regular files
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 * - shared mappings on character devices and memory backed files are
1262 * permitted to overlap inexactly as far as we are concerned for in
1263 * these cases, sharing is handled in the driver or filesystem rather
1264 * than here
1265 */
1266 if (vm_flags & VM_MAYSHARE) {
David Howells8feae132009-01-08 12:04:47 +00001267 struct vm_region *pregion;
1268 unsigned long pglen, rpglen, pgend, rpgend, start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269
David Howells8feae132009-01-08 12:04:47 +00001270 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1271 pgend = pgoff + pglen;
David Howells165b2392007-03-22 00:11:24 -08001272
David Howells8feae132009-01-08 12:04:47 +00001273 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1274 pregion = rb_entry(rb, struct vm_region, vm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275
David Howells8feae132009-01-08 12:04:47 +00001276 if (!(pregion->vm_flags & VM_MAYSHARE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 continue;
1278
1279 /* search for overlapping mappings on the same file */
Al Viro496ad9a2013-01-23 17:07:38 -05001280 if (file_inode(pregion->vm_file) !=
1281 file_inode(file))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 continue;
1283
David Howells8feae132009-01-08 12:04:47 +00001284 if (pregion->vm_pgoff >= pgend)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 continue;
1286
David Howells8feae132009-01-08 12:04:47 +00001287 rpglen = pregion->vm_end - pregion->vm_start;
1288 rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1289 rpgend = pregion->vm_pgoff + rpglen;
1290 if (pgoff >= rpgend)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291 continue;
1292
David Howells8feae132009-01-08 12:04:47 +00001293 /* handle inexactly overlapping matches between
1294 * mappings */
1295 if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1296 !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1297 /* new mapping is not a subset of the region */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001298 if (!(capabilities & NOMMU_MAP_DIRECT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 goto sharing_violation;
1300 continue;
1301 }
1302
David Howells8feae132009-01-08 12:04:47 +00001303 /* we've found a region we can share */
David Howells1e2ae592010-01-15 17:01:33 -08001304 pregion->vm_usage++;
David Howells8feae132009-01-08 12:04:47 +00001305 vma->vm_region = pregion;
1306 start = pregion->vm_start;
1307 start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1308 vma->vm_start = start;
1309 vma->vm_end = start + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001311 if (pregion->vm_flags & VM_MAPPED_COPY)
David Howells8feae132009-01-08 12:04:47 +00001312 vma->vm_flags |= VM_MAPPED_COPY;
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001313 else {
David Howells8feae132009-01-08 12:04:47 +00001314 ret = do_mmap_shared_file(vma);
1315 if (ret < 0) {
1316 vma->vm_region = NULL;
1317 vma->vm_start = 0;
1318 vma->vm_end = 0;
David Howells1e2ae592010-01-15 17:01:33 -08001319 pregion->vm_usage--;
David Howells8feae132009-01-08 12:04:47 +00001320 pregion = NULL;
1321 goto error_just_free;
1322 }
1323 }
1324 fput(region->vm_file);
1325 kmem_cache_free(vm_region_jar, region);
1326 region = pregion;
1327 result = start;
1328 goto share;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329 }
1330
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331 /* obtain the address at which to make a shared mapping
1332 * - this is the hook for quasi-memory character devices to
1333 * tell us the location of a shared mapping
1334 */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001335 if (capabilities & NOMMU_MAP_DIRECT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 addr = file->f_op->get_unmapped_area(file, addr, len,
1337 pgoff, flags);
Namhyung Kimbb005a52011-05-24 17:11:27 -07001338 if (IS_ERR_VALUE(addr)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 ret = addr;
Namhyung Kimbb005a52011-05-24 17:11:27 -07001340 if (ret != -ENOSYS)
David Howells8feae132009-01-08 12:04:47 +00001341 goto error_just_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342
1343 /* the driver refused to tell us where to site
1344 * the mapping so we'll have to attempt to copy
1345 * it */
Namhyung Kimbb005a52011-05-24 17:11:27 -07001346 ret = -ENODEV;
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001347 if (!(capabilities & NOMMU_MAP_COPY))
David Howells8feae132009-01-08 12:04:47 +00001348 goto error_just_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001350 capabilities &= ~NOMMU_MAP_DIRECT;
David Howells8feae132009-01-08 12:04:47 +00001351 } else {
1352 vma->vm_start = region->vm_start = addr;
1353 vma->vm_end = region->vm_end = addr + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 }
1355 }
1356 }
1357
David Howells8feae132009-01-08 12:04:47 +00001358 vma->vm_region = region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359
David Howells645d83c2009-09-24 15:13:10 +01001360 /* set up the mapping
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001361 * - the region is filled in if NOMMU_MAP_DIRECT is still set
David Howells645d83c2009-09-24 15:13:10 +01001362 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 if (file && vma->vm_flags & VM_SHARED)
David Howells8feae132009-01-08 12:04:47 +00001364 ret = do_mmap_shared_file(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 else
David Howells645d83c2009-09-24 15:13:10 +01001366 ret = do_mmap_private(vma, region, len, capabilities);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367 if (ret < 0)
David Howells645d83c2009-09-24 15:13:10 +01001368 goto error_just_free;
1369 add_nommu_region(region);
David Howells8feae132009-01-08 12:04:47 +00001370
Jie Zhangea637632009-12-14 18:00:02 -08001371 /* clear anonymous mappings that don't ask for uninitialized data */
1372 if (!vma->vm_file && !(flags & MAP_UNINITIALIZED))
1373 memset((void *)region->vm_start, 0,
1374 region->vm_end - region->vm_start);
1375
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 /* okay... we have a mapping; now we have to register it */
David Howells8feae132009-01-08 12:04:47 +00001377 result = vma->vm_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 current->mm->total_vm += len >> PAGE_SHIFT;
1380
David Howells8feae132009-01-08 12:04:47 +00001381share:
1382 add_vma_to_mm(current->mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383
Mike Frysingercfe79c02010-01-06 17:23:23 +00001384 /* we flush the region from the icache only when the first executable
1385 * mapping of it is made */
1386 if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1387 flush_icache_range(region->vm_start, region->vm_end);
1388 region->vm_icache_flushed = true;
1389 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390
Mike Frysingercfe79c02010-01-06 17:23:23 +00001391 up_write(&nommu_region_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392
David Howells8feae132009-01-08 12:04:47 +00001393 return result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394
David Howells8feae132009-01-08 12:04:47 +00001395error_just_free:
1396 up_write(&nommu_region_sem);
1397error:
David Howells89a86402009-10-30 13:13:26 +00001398 if (region->vm_file)
1399 fput(region->vm_file);
David Howells8feae132009-01-08 12:04:47 +00001400 kmem_cache_free(vm_region_jar, region);
David Howells89a86402009-10-30 13:13:26 +00001401 if (vma->vm_file)
1402 fput(vma->vm_file);
David Howells8feae132009-01-08 12:04:47 +00001403 kmem_cache_free(vm_area_cachep, vma);
David Howells8feae132009-01-08 12:04:47 +00001404 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405
David Howells8feae132009-01-08 12:04:47 +00001406sharing_violation:
1407 up_write(&nommu_region_sem);
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001408 pr_warn("Attempt to share mismatched mappings\n");
David Howells8feae132009-01-08 12:04:47 +00001409 ret = -EINVAL;
1410 goto error;
1411
1412error_getting_vma:
1413 kmem_cache_free(vm_region_jar, region);
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001414 pr_warn("Allocation of vma for %lu byte allocation from process %d failed\n",
1415 len, current->pid);
Michal Hocko9af744d2017-02-22 15:46:16 -08001416 show_free_areas(0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 return -ENOMEM;
1418
David Howells8feae132009-01-08 12:04:47 +00001419error_getting_region:
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001420 pr_warn("Allocation of vm region for %lu byte allocation from process %d failed\n",
1421 len, current->pid);
Michal Hocko9af744d2017-02-22 15:46:16 -08001422 show_free_areas(0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 return -ENOMEM;
1424}
Linus Torvalds6be5ceb2012-04-20 17:13:58 -07001425
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001426SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1427 unsigned long, prot, unsigned long, flags,
1428 unsigned long, fd, unsigned long, pgoff)
1429{
1430 struct file *file = NULL;
1431 unsigned long retval = -EBADF;
1432
Al Viro120a7952010-10-30 02:54:44 -04001433 audit_mmap_fd(fd, flags);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001434 if (!(flags & MAP_ANONYMOUS)) {
1435 file = fget(fd);
1436 if (!file)
1437 goto out;
1438 }
1439
1440 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1441
Greg Ungererad1ed292012-06-04 14:29:59 +10001442 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001443
1444 if (file)
1445 fput(file);
1446out:
1447 return retval;
1448}
1449
Christoph Hellwiga4679372010-03-10 15:21:15 -08001450#ifdef __ARCH_WANT_SYS_OLD_MMAP
1451struct mmap_arg_struct {
1452 unsigned long addr;
1453 unsigned long len;
1454 unsigned long prot;
1455 unsigned long flags;
1456 unsigned long fd;
1457 unsigned long offset;
1458};
1459
1460SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1461{
1462 struct mmap_arg_struct a;
1463
1464 if (copy_from_user(&a, arg, sizeof(a)))
1465 return -EFAULT;
Alexander Kuleshov1824cb72015-11-05 18:46:35 -08001466 if (offset_in_page(a.offset))
Christoph Hellwiga4679372010-03-10 15:21:15 -08001467 return -EINVAL;
1468
1469 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1470 a.offset >> PAGE_SHIFT);
1471}
1472#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1473
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474/*
David Howells8feae132009-01-08 12:04:47 +00001475 * split a vma into two pieces at address 'addr', a new vma is allocated either
1476 * for the first part or the tail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 */
David Howells8feae132009-01-08 12:04:47 +00001478int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1479 unsigned long addr, int new_below)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480{
David Howells8feae132009-01-08 12:04:47 +00001481 struct vm_area_struct *new;
1482 struct vm_region *region;
1483 unsigned long npages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484
David Howells779c1022010-01-15 17:01:34 -08001485 /* we're only permitted to split anonymous regions (these should have
1486 * only a single usage on the region) */
1487 if (vma->vm_file)
David Howells8feae132009-01-08 12:04:47 +00001488 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489
David Howells8feae132009-01-08 12:04:47 +00001490 if (mm->map_count >= sysctl_max_map_count)
1491 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492
David Howells8feae132009-01-08 12:04:47 +00001493 region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1494 if (!region)
1495 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496
David Howells8feae132009-01-08 12:04:47 +00001497 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1498 if (!new) {
1499 kmem_cache_free(vm_region_jar, region);
1500 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501 }
David Howells8feae132009-01-08 12:04:47 +00001502
1503 /* most fields are the same, copy all, and then fixup */
1504 *new = *vma;
1505 *region = *vma->vm_region;
1506 new->vm_region = region;
1507
1508 npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1509
1510 if (new_below) {
Paul Mundtdd8632a2009-01-08 12:04:47 +00001511 region->vm_top = region->vm_end = new->vm_end = addr;
David Howells8feae132009-01-08 12:04:47 +00001512 } else {
1513 region->vm_start = new->vm_start = addr;
1514 region->vm_pgoff = new->vm_pgoff += npages;
1515 }
1516
1517 if (new->vm_ops && new->vm_ops->open)
1518 new->vm_ops->open(new);
1519
1520 delete_vma_from_mm(vma);
1521 down_write(&nommu_region_sem);
1522 delete_nommu_region(vma->vm_region);
1523 if (new_below) {
1524 vma->vm_region->vm_start = vma->vm_start = addr;
1525 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1526 } else {
1527 vma->vm_region->vm_end = vma->vm_end = addr;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001528 vma->vm_region->vm_top = addr;
David Howells8feae132009-01-08 12:04:47 +00001529 }
1530 add_nommu_region(vma->vm_region);
1531 add_nommu_region(new->vm_region);
1532 up_write(&nommu_region_sem);
1533 add_vma_to_mm(mm, vma);
1534 add_vma_to_mm(mm, new);
1535 return 0;
1536}
1537
1538/*
1539 * shrink a VMA by removing the specified chunk from either the beginning or
1540 * the end
1541 */
1542static int shrink_vma(struct mm_struct *mm,
1543 struct vm_area_struct *vma,
1544 unsigned long from, unsigned long to)
1545{
1546 struct vm_region *region;
1547
David Howells8feae132009-01-08 12:04:47 +00001548 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1549 * and list */
1550 delete_vma_from_mm(vma);
1551 if (from > vma->vm_start)
1552 vma->vm_end = from;
1553 else
1554 vma->vm_start = to;
1555 add_vma_to_mm(mm, vma);
1556
1557 /* cut the backing region down to size */
1558 region = vma->vm_region;
David Howells1e2ae592010-01-15 17:01:33 -08001559 BUG_ON(region->vm_usage != 1);
David Howells8feae132009-01-08 12:04:47 +00001560
1561 down_write(&nommu_region_sem);
1562 delete_nommu_region(region);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001563 if (from > region->vm_start) {
1564 to = region->vm_top;
1565 region->vm_top = region->vm_end = from;
1566 } else {
David Howells8feae132009-01-08 12:04:47 +00001567 region->vm_start = to;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001568 }
David Howells8feae132009-01-08 12:04:47 +00001569 add_nommu_region(region);
1570 up_write(&nommu_region_sem);
1571
1572 free_page_series(from, to);
1573 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574}
1575
David Howells30340972006-09-27 01:50:20 -07001576/*
1577 * release a mapping
David Howells8feae132009-01-08 12:04:47 +00001578 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1579 * VMA, though it need not cover the whole VMA
David Howells30340972006-09-27 01:50:20 -07001580 */
Mike Rapoport897ab3e2017-02-24 14:58:22 -08001581int do_munmap(struct mm_struct *mm, unsigned long start, size_t len, struct list_head *uf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582{
David Howells8feae132009-01-08 12:04:47 +00001583 struct vm_area_struct *vma;
Bob Liuf67d9b12011-05-24 17:12:56 -07001584 unsigned long end;
David Howells8feae132009-01-08 12:04:47 +00001585 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586
Bob Liuf67d9b12011-05-24 17:12:56 -07001587 len = PAGE_ALIGN(len);
David Howells8feae132009-01-08 12:04:47 +00001588 if (len == 0)
1589 return -EINVAL;
1590
Bob Liuf67d9b12011-05-24 17:12:56 -07001591 end = start + len;
1592
David Howells8feae132009-01-08 12:04:47 +00001593 /* find the first potentially overlapping VMA */
1594 vma = find_vma(mm, start);
1595 if (!vma) {
Choi Gi-yongac714902014-04-07 15:37:36 -07001596 static int limit;
David Howells33e5d7692009-04-02 16:56:32 -07001597 if (limit < 5) {
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001598 pr_warn("munmap of memory not mmapped by process %d (%s): 0x%lx-0x%lx\n",
1599 current->pid, current->comm,
1600 start, start + len - 1);
David Howells33e5d7692009-04-02 16:56:32 -07001601 limit++;
1602 }
David Howells8feae132009-01-08 12:04:47 +00001603 return -EINVAL;
David Howells30340972006-09-27 01:50:20 -07001604 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605
David Howells8feae132009-01-08 12:04:47 +00001606 /* we're allowed to split an anonymous VMA but not a file-backed one */
1607 if (vma->vm_file) {
1608 do {
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001609 if (start > vma->vm_start)
David Howells8feae132009-01-08 12:04:47 +00001610 return -EINVAL;
David Howells8feae132009-01-08 12:04:47 +00001611 if (end == vma->vm_end)
1612 goto erase_whole_vma;
Namhyung Kimd75a3102011-05-24 17:11:25 -07001613 vma = vma->vm_next;
1614 } while (vma);
David Howells8feae132009-01-08 12:04:47 +00001615 return -EINVAL;
1616 } else {
1617 /* the chunk must be a subset of the VMA found */
1618 if (start == vma->vm_start && end == vma->vm_end)
1619 goto erase_whole_vma;
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001620 if (start < vma->vm_start || end > vma->vm_end)
David Howells8feae132009-01-08 12:04:47 +00001621 return -EINVAL;
Alexander Kuleshov1824cb72015-11-05 18:46:35 -08001622 if (offset_in_page(start))
David Howells8feae132009-01-08 12:04:47 +00001623 return -EINVAL;
Alexander Kuleshov1824cb72015-11-05 18:46:35 -08001624 if (end != vma->vm_end && offset_in_page(end))
David Howells8feae132009-01-08 12:04:47 +00001625 return -EINVAL;
David Howells8feae132009-01-08 12:04:47 +00001626 if (start != vma->vm_start && end != vma->vm_end) {
1627 ret = split_vma(mm, vma, start, 1);
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001628 if (ret < 0)
David Howells8feae132009-01-08 12:04:47 +00001629 return ret;
David Howells8feae132009-01-08 12:04:47 +00001630 }
1631 return shrink_vma(mm, vma, start, end);
1632 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633
David Howells8feae132009-01-08 12:04:47 +00001634erase_whole_vma:
1635 delete_vma_from_mm(vma);
1636 delete_vma(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 return 0;
1638}
Paul Mundtb5073172007-07-21 04:37:25 -07001639EXPORT_SYMBOL(do_munmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640
Al Virobfce2812012-04-20 21:57:04 -04001641int vm_munmap(unsigned long addr, size_t len)
David Howells30340972006-09-27 01:50:20 -07001642{
Al Virobfce2812012-04-20 21:57:04 -04001643 struct mm_struct *mm = current->mm;
David Howells30340972006-09-27 01:50:20 -07001644 int ret;
David Howells30340972006-09-27 01:50:20 -07001645
1646 down_write(&mm->mmap_sem);
Mike Rapoport897ab3e2017-02-24 14:58:22 -08001647 ret = do_munmap(mm, addr, len, NULL);
David Howells30340972006-09-27 01:50:20 -07001648 up_write(&mm->mmap_sem);
1649 return ret;
1650}
Linus Torvaldsa46ef992012-04-20 16:20:01 -07001651EXPORT_SYMBOL(vm_munmap);
1652
1653SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1654{
Al Virobfce2812012-04-20 21:57:04 -04001655 return vm_munmap(addr, len);
Linus Torvaldsa46ef992012-04-20 16:20:01 -07001656}
David Howells30340972006-09-27 01:50:20 -07001657
1658/*
David Howells8feae132009-01-08 12:04:47 +00001659 * release all the mappings made in a process's VM space
David Howells30340972006-09-27 01:50:20 -07001660 */
David Howells8feae132009-01-08 12:04:47 +00001661void exit_mmap(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662{
David Howells8feae132009-01-08 12:04:47 +00001663 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664
David Howells8feae132009-01-08 12:04:47 +00001665 if (!mm)
1666 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667
David Howells8feae132009-01-08 12:04:47 +00001668 mm->total_vm = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669
David Howells8feae132009-01-08 12:04:47 +00001670 while ((vma = mm->mmap)) {
1671 mm->mmap = vma->vm_next;
1672 delete_vma_from_mm(vma);
1673 delete_vma(mm, vma);
Steven J. Magnani04c34962010-11-24 12:56:54 -08001674 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675 }
1676}
1677
Linus Torvalds5d22fc22016-05-27 15:57:31 -07001678int vm_brk(unsigned long addr, unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679{
1680 return -ENOMEM;
1681}
1682
1683/*
David Howells6fa5f802006-09-27 01:50:21 -07001684 * expand (or shrink) an existing mapping, potentially moving it at the same
1685 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686 *
David Howells6fa5f802006-09-27 01:50:21 -07001687 * under NOMMU conditions, we only permit changing a mapping's size, and only
David Howells8feae132009-01-08 12:04:47 +00001688 * as long as it stays within the region allocated by do_mmap_private() and the
1689 * block is not shareable
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690 *
David Howells6fa5f802006-09-27 01:50:21 -07001691 * MREMAP_FIXED is not supported under NOMMU conditions
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692 */
Al Viro4b377ba2013-03-04 10:47:59 -05001693static unsigned long do_mremap(unsigned long addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694 unsigned long old_len, unsigned long new_len,
1695 unsigned long flags, unsigned long new_addr)
1696{
David Howells6fa5f802006-09-27 01:50:21 -07001697 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698
1699 /* insanity checks first */
Bob Liuf67d9b12011-05-24 17:12:56 -07001700 old_len = PAGE_ALIGN(old_len);
1701 new_len = PAGE_ALIGN(new_len);
David Howells8feae132009-01-08 12:04:47 +00001702 if (old_len == 0 || new_len == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703 return (unsigned long) -EINVAL;
1704
Alexander Kuleshov1824cb72015-11-05 18:46:35 -08001705 if (offset_in_page(addr))
David Howells8feae132009-01-08 12:04:47 +00001706 return -EINVAL;
1707
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708 if (flags & MREMAP_FIXED && new_addr != addr)
1709 return (unsigned long) -EINVAL;
1710
David Howells8feae132009-01-08 12:04:47 +00001711 vma = find_vma_exact(current->mm, addr, old_len);
David Howells6fa5f802006-09-27 01:50:21 -07001712 if (!vma)
1713 return (unsigned long) -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714
David Howells6fa5f802006-09-27 01:50:21 -07001715 if (vma->vm_end != vma->vm_start + old_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716 return (unsigned long) -EFAULT;
1717
David Howells6fa5f802006-09-27 01:50:21 -07001718 if (vma->vm_flags & VM_MAYSHARE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719 return (unsigned long) -EPERM;
1720
David Howells8feae132009-01-08 12:04:47 +00001721 if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722 return (unsigned long) -ENOMEM;
1723
1724 /* all checks complete - do it */
David Howells6fa5f802006-09-27 01:50:21 -07001725 vma->vm_end = vma->vm_start + new_len;
David Howells6fa5f802006-09-27 01:50:21 -07001726 return vma->vm_start;
1727}
1728
Heiko Carstens6a6160a2009-01-14 14:14:15 +01001729SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1730 unsigned long, new_len, unsigned long, flags,
1731 unsigned long, new_addr)
David Howells6fa5f802006-09-27 01:50:21 -07001732{
1733 unsigned long ret;
1734
1735 down_write(&current->mm->mmap_sem);
1736 ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1737 up_write(&current->mm->mmap_sem);
1738 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739}
1740
Michel Lespinasse240aade2013-02-22 16:35:56 -08001741struct page *follow_page_mask(struct vm_area_struct *vma,
1742 unsigned long address, unsigned int flags,
1743 unsigned int *page_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744{
Michel Lespinasse240aade2013-02-22 16:35:56 -08001745 *page_mask = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746 return NULL;
1747}
1748
Bob Liu8f3b1322011-07-08 15:39:46 -07001749int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1750 unsigned long pfn, unsigned long size, pgprot_t prot)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751{
Bob Liu8f3b1322011-07-08 15:39:46 -07001752 if (addr != (pfn << PAGE_SHIFT))
1753 return -EINVAL;
1754
Konstantin Khlebnikov314e51b2012-10-08 16:29:02 -07001755 vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
Greg Ungerer66aa2b42005-09-12 11:18:10 +10001756 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757}
Luke Yang22c4af42006-07-14 00:24:09 -07001758EXPORT_SYMBOL(remap_pfn_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759
Linus Torvalds3c0b9de2013-04-27 13:25:38 -07001760int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)
1761{
1762 unsigned long pfn = start >> PAGE_SHIFT;
1763 unsigned long vm_len = vma->vm_end - vma->vm_start;
1764
1765 pfn += vma->vm_pgoff;
1766 return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);
1767}
1768EXPORT_SYMBOL(vm_iomap_memory);
1769
Paul Mundtf905bc42008-02-04 22:29:59 -08001770int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1771 unsigned long pgoff)
1772{
1773 unsigned int size = vma->vm_end - vma->vm_start;
1774
1775 if (!(vma->vm_flags & VM_USERMAP))
1776 return -EINVAL;
1777
1778 vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1779 vma->vm_end = vma->vm_start + size;
1780
1781 return 0;
1782}
1783EXPORT_SYMBOL(remap_vmalloc_range);
1784
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1786 unsigned long len, unsigned long pgoff, unsigned long flags)
1787{
1788 return -ENOMEM;
1789}
1790
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791void unmap_mapping_range(struct address_space *mapping,
1792 loff_t const holebegin, loff_t const holelen,
1793 int even_cows)
1794{
1795}
Luke Yang22c4af42006-07-14 00:24:09 -07001796EXPORT_SYMBOL(unmap_mapping_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797
Dave Jiang11bac802017-02-24 14:56:41 -08001798int filemap_fault(struct vm_fault *vmf)
David Howellsb0e15192006-01-06 00:11:42 -08001799{
1800 BUG();
Nick Piggind0217ac2007-07-19 01:47:03 -07001801 return 0;
David Howellsb0e15192006-01-06 00:11:42 -08001802}
Paul Mundtb5073172007-07-21 04:37:25 -07001803EXPORT_SYMBOL(filemap_fault);
David Howells0ec76a12006-09-27 01:50:15 -07001804
Jan Kara82b0f8c2016-12-14 15:06:58 -08001805void filemap_map_pages(struct vm_fault *vmf,
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07001806 pgoff_t start_pgoff, pgoff_t end_pgoff)
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07001807{
1808 BUG();
1809}
1810EXPORT_SYMBOL(filemap_map_pages);
1811
Eric W. Biederman84d77d32016-11-22 12:06:50 -06001812int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
Lorenzo Stoakes442486e2016-10-13 01:20:18 +01001813 unsigned long addr, void *buf, int len, unsigned int gup_flags)
David Howells0ec76a12006-09-27 01:50:15 -07001814{
David Howells0ec76a12006-09-27 01:50:15 -07001815 struct vm_area_struct *vma;
Lorenzo Stoakes442486e2016-10-13 01:20:18 +01001816 int write = gup_flags & FOLL_WRITE;
David Howells0ec76a12006-09-27 01:50:15 -07001817
1818 down_read(&mm->mmap_sem);
1819
1820 /* the access must start within one of the target process's mappings */
David Howells0159b142006-09-27 01:50:16 -07001821 vma = find_vma(mm, addr);
1822 if (vma) {
David Howells0ec76a12006-09-27 01:50:15 -07001823 /* don't overrun this mapping */
1824 if (addr + len >= vma->vm_end)
1825 len = vma->vm_end - addr;
1826
1827 /* only read or write mappings where it is permitted */
David Howellsd00c7b992006-09-27 01:50:19 -07001828 if (write && vma->vm_flags & VM_MAYWRITE)
Jie Zhang79597222010-01-06 17:23:28 +00001829 copy_to_user_page(vma, NULL, addr,
1830 (void *) addr, buf, len);
David Howellsd00c7b992006-09-27 01:50:19 -07001831 else if (!write && vma->vm_flags & VM_MAYREAD)
Jie Zhang79597222010-01-06 17:23:28 +00001832 copy_from_user_page(vma, NULL, addr,
1833 buf, (void *) addr, len);
David Howells0ec76a12006-09-27 01:50:15 -07001834 else
1835 len = 0;
1836 } else {
1837 len = 0;
1838 }
1839
1840 up_read(&mm->mmap_sem);
Mike Frysingerf55f1992011-03-29 14:05:12 +01001841
1842 return len;
1843}
1844
1845/**
1846 * @access_remote_vm - access another process' address space
1847 * @mm: the mm_struct of the target address space
1848 * @addr: start address to access
1849 * @buf: source or destination buffer
1850 * @len: number of bytes to transfer
Lorenzo Stoakes6347e8d2016-10-13 01:20:19 +01001851 * @gup_flags: flags modifying lookup behaviour
Mike Frysingerf55f1992011-03-29 14:05:12 +01001852 *
1853 * The caller must hold a reference on @mm.
1854 */
1855int access_remote_vm(struct mm_struct *mm, unsigned long addr,
Lorenzo Stoakes6347e8d2016-10-13 01:20:19 +01001856 void *buf, int len, unsigned int gup_flags)
Mike Frysingerf55f1992011-03-29 14:05:12 +01001857{
Lorenzo Stoakes6347e8d2016-10-13 01:20:19 +01001858 return __access_remote_vm(NULL, mm, addr, buf, len, gup_flags);
Mike Frysingerf55f1992011-03-29 14:05:12 +01001859}
1860
1861/*
1862 * Access another process' address space.
1863 * - source/target buffer must be kernel space
1864 */
Lorenzo Stoakesf307ab62016-10-13 01:20:20 +01001865int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len,
1866 unsigned int gup_flags)
Mike Frysingerf55f1992011-03-29 14:05:12 +01001867{
1868 struct mm_struct *mm;
1869
1870 if (addr + len < addr)
1871 return 0;
1872
1873 mm = get_task_mm(tsk);
1874 if (!mm)
1875 return 0;
1876
Lorenzo Stoakesf307ab62016-10-13 01:20:20 +01001877 len = __access_remote_vm(tsk, mm, addr, buf, len, gup_flags);
Mike Frysingerf55f1992011-03-29 14:05:12 +01001878
David Howells0ec76a12006-09-27 01:50:15 -07001879 mmput(mm);
1880 return len;
1881}
Catalin Marinasfcd35852016-11-01 14:43:25 -07001882EXPORT_SYMBOL_GPL(access_process_vm);
David Howells7e660872010-01-15 17:01:39 -08001883
1884/**
1885 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
1886 * @inode: The inode to check
1887 * @size: The current filesize of the inode
1888 * @newsize: The proposed filesize of the inode
1889 *
1890 * Check the shared mappings on an inode on behalf of a shrinking truncate to
1891 * make sure that that any outstanding VMAs aren't broken and then shrink the
1892 * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
1893 * automatically grant mappings that are too large.
1894 */
1895int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
1896 size_t newsize)
1897{
1898 struct vm_area_struct *vma;
David Howells7e660872010-01-15 17:01:39 -08001899 struct vm_region *region;
1900 pgoff_t low, high;
1901 size_t r_size, r_top;
1902
1903 low = newsize >> PAGE_SHIFT;
1904 high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1905
1906 down_write(&nommu_region_sem);
Davidlohr Bueso1acf2e02014-12-12 16:54:39 -08001907 i_mmap_lock_read(inode->i_mapping);
David Howells7e660872010-01-15 17:01:39 -08001908
1909 /* search for VMAs that fall within the dead zone */
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -07001910 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) {
David Howells7e660872010-01-15 17:01:39 -08001911 /* found one - only interested if it's shared out of the page
1912 * cache */
1913 if (vma->vm_flags & VM_SHARED) {
Davidlohr Bueso1acf2e02014-12-12 16:54:39 -08001914 i_mmap_unlock_read(inode->i_mapping);
David Howells7e660872010-01-15 17:01:39 -08001915 up_write(&nommu_region_sem);
1916 return -ETXTBSY; /* not quite true, but near enough */
1917 }
1918 }
1919
1920 /* reduce any regions that overlap the dead zone - if in existence,
1921 * these will be pointed to by VMAs that don't overlap the dead zone
1922 *
1923 * we don't check for any regions that start beyond the EOF as there
1924 * shouldn't be any
1925 */
Davidlohr Bueso1acf2e02014-12-12 16:54:39 -08001926 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, 0, ULONG_MAX) {
David Howells7e660872010-01-15 17:01:39 -08001927 if (!(vma->vm_flags & VM_SHARED))
1928 continue;
1929
1930 region = vma->vm_region;
1931 r_size = region->vm_top - region->vm_start;
1932 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
1933
1934 if (r_top > newsize) {
1935 region->vm_top -= r_top - newsize;
1936 if (region->vm_end > region->vm_top)
1937 region->vm_end = region->vm_top;
1938 }
1939 }
1940
Davidlohr Bueso1acf2e02014-12-12 16:54:39 -08001941 i_mmap_unlock_read(inode->i_mapping);
David Howells7e660872010-01-15 17:01:39 -08001942 up_write(&nommu_region_sem);
1943 return 0;
1944}
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07001945
1946/*
1947 * Initialise sysctl_user_reserve_kbytes.
1948 *
1949 * This is intended to prevent a user from starting a single memory hogging
1950 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
1951 * mode.
1952 *
1953 * The default value is min(3% of free memory, 128MB)
1954 * 128MB is enough to recover with sshd/login, bash, and top/kill.
1955 */
1956static int __meminit init_user_reserve(void)
1957{
1958 unsigned long free_kbytes;
1959
Michal Hockoc41f0122017-09-06 16:23:36 -07001960 free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07001961
1962 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
1963 return 0;
1964}
Paul Gortmakera4bc6fc2015-05-01 20:08:20 -04001965subsys_initcall(init_user_reserve);
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07001966
1967/*
1968 * Initialise sysctl_admin_reserve_kbytes.
1969 *
1970 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
1971 * to log in and kill a memory hogging process.
1972 *
1973 * Systems with more than 256MB will reserve 8MB, enough to recover
1974 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
1975 * only reserve 3% of free pages by default.
1976 */
1977static int __meminit init_admin_reserve(void)
1978{
1979 unsigned long free_kbytes;
1980
Michal Hockoc41f0122017-09-06 16:23:36 -07001981 free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07001982
1983 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
1984 return 0;
1985}
Paul Gortmakera4bc6fc2015-05-01 20:08:20 -04001986subsys_initcall(init_admin_reserve);