blob: b40ec74f364c547de764d64bce43f0f09f52bd0d [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>
Davidlohr Bueso615d6e82014-04-07 15:37:25 -070020#include <linux/vmacache.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include <linux/mman.h>
22#include <linux/swap.h>
23#include <linux/file.h>
24#include <linux/highmem.h>
25#include <linux/pagemap.h>
26#include <linux/slab.h>
27#include <linux/vmalloc.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <linux/blkdev.h>
29#include <linux/backing-dev.h>
Gideon Israel Dsouza3b321232014-04-07 15:37:26 -070030#include <linux/compiler.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include <linux/mount.h>
32#include <linux/personality.h>
33#include <linux/security.h>
34#include <linux/syscalls.h>
Al Viro120a7952010-10-30 02:54:44 -040035#include <linux/audit.h>
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -070036#include <linux/printk.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
38#include <asm/uaccess.h>
39#include <asm/tlb.h>
40#include <asm/tlbflush.h>
Bernd Schmidteb8cdec2009-09-21 17:03:57 -070041#include <asm/mmu_context.h>
David Howells8feae132009-01-08 12:04:47 +000042#include "internal.h"
43
Linus Torvalds1da177e2005-04-16 15:20:36 -070044void *high_memory;
Arnd Bergmann944b6872015-02-05 12:25:12 -080045EXPORT_SYMBOL(high_memory);
Linus Torvalds1da177e2005-04-16 15:20:36 -070046struct page *mem_map;
47unsigned long max_mapnr;
gchen gchen5b8bf302015-03-12 16:26:05 -070048EXPORT_SYMBOL(max_mapnr);
Hugh Dickins4266c972009-09-23 17:05:53 +010049unsigned long highest_memmap_pfn;
David Howellsfc4d5c22009-05-06 16:03:05 -070050int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -070051int heap_stack_gap = 0;
52
David Howells33e5d7692009-04-02 16:56:32 -070053atomic_long_t mmap_pages_allocated;
David Howells8feae132009-01-08 12:04:47 +000054
Linus Torvalds1da177e2005-04-16 15:20:36 -070055EXPORT_SYMBOL(mem_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -070056
David Howells8feae132009-01-08 12:04:47 +000057/* list of mapped, potentially shareable regions */
58static struct kmem_cache *vm_region_jar;
59struct rb_root nommu_region_tree = RB_ROOT;
60DECLARE_RWSEM(nommu_region_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -070061
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +040062const struct vm_operations_struct generic_file_vm_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -070063};
64
65/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070066 * Return the total memory allocated for this pointer, not
67 * just what the caller asked for.
68 *
69 * Doesn't have to be accurate, i.e. may have races.
70 */
71unsigned int kobjsize(const void *objp)
72{
73 struct page *page;
74
Michael Hennerich4016a132008-04-28 02:13:38 -070075 /*
76 * If the object we have should not have ksize performed on it,
77 * return size of 0
78 */
Paul Mundt5a1603b2008-06-12 16:29:55 +090079 if (!objp || !virt_addr_valid(objp))
Paul Mundt6cfd53fc2008-06-05 22:46:08 -070080 return 0;
81
82 page = virt_to_head_page(objp);
Paul Mundt6cfd53fc2008-06-05 22:46:08 -070083
84 /*
85 * If the allocator sets PageSlab, we know the pointer came from
86 * kmalloc().
87 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070088 if (PageSlab(page))
89 return ksize(objp);
90
Paul Mundt6cfd53fc2008-06-05 22:46:08 -070091 /*
Paul Mundtab2e83e2009-01-08 12:04:48 +000092 * If it's not a compound page, see if we have a matching VMA
93 * region. This test is intentionally done in reverse order,
94 * so if there's no VMA, we still fall through and hand back
95 * PAGE_SIZE for 0-order pages.
96 */
97 if (!PageCompound(page)) {
98 struct vm_area_struct *vma;
99
100 vma = find_vma(current->mm, (unsigned long)objp);
101 if (vma)
102 return vma->vm_end - vma->vm_start;
103 }
104
105 /*
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700106 * The ksize() function is only guaranteed to work for pointers
Paul Mundt5a1603b2008-06-12 16:29:55 +0900107 * returned by kmalloc(). So handle arbitrary pointers here.
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700108 */
Paul Mundt5a1603b2008-06-12 16:29:55 +0900109 return PAGE_SIZE << compound_order(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110}
111
Lorenzo Stoakes0d731752016-10-24 10:57:25 +0100112static long __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
Michel Lespinasse28a35712013-02-22 16:35:55 -0800113 unsigned long start, unsigned long nr_pages,
114 unsigned int foll_flags, struct page **pages,
115 struct vm_area_struct **vmas, int *nonblocking)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116{
Sonic Zhang910e46d2006-09-27 01:50:17 -0700117 struct vm_area_struct *vma;
David Howells7b4d5b82006-09-27 01:50:18 -0700118 unsigned long vm_flags;
119 int i;
120
121 /* calculate required read or write permissions.
Hugh Dickins58fa8792009-09-21 17:03:31 -0700122 * If FOLL_FORCE is set, we only require the "MAY" flags.
David Howells7b4d5b82006-09-27 01:50:18 -0700123 */
Hugh Dickins58fa8792009-09-21 17:03:31 -0700124 vm_flags = (foll_flags & FOLL_WRITE) ?
125 (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
126 vm_flags &= (foll_flags & FOLL_FORCE) ?
127 (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128
Peter Zijlstra9d737772009-06-25 11:58:55 +0200129 for (i = 0; i < nr_pages; i++) {
David Howells7561e8c2010-03-25 16:48:38 +0000130 vma = find_vma(mm, start);
David Howells7b4d5b82006-09-27 01:50:18 -0700131 if (!vma)
132 goto finish_or_fault;
133
134 /* protect what we can, including chardevs */
Hugh Dickins1c3aff12009-09-21 17:03:24 -0700135 if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
136 !(vm_flags & vma->vm_flags))
David Howells7b4d5b82006-09-27 01:50:18 -0700137 goto finish_or_fault;
Sonic Zhang910e46d2006-09-27 01:50:17 -0700138
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 if (pages) {
140 pages[i] = virt_to_page(start);
141 if (pages[i])
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300142 get_page(pages[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143 }
144 if (vmas)
Sonic Zhang910e46d2006-09-27 01:50:17 -0700145 vmas[i] = vma;
David Howellse1ee65d2010-03-25 16:48:44 +0000146 start = (start + PAGE_SIZE) & PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 }
David Howells7b4d5b82006-09-27 01:50:18 -0700148
149 return i;
150
151finish_or_fault:
152 return i ? : -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153}
Nick Pigginb291f002008-10-18 20:26:44 -0700154
Nick Pigginb291f002008-10-18 20:26:44 -0700155/*
156 * get a list of pages in an address range belonging to the specified process
157 * and indicate the VMA that covers each page
158 * - this is potentially dodgy as we may end incrementing the page count of a
159 * slab page or a secondary page from a compound page
160 * - don't permit access to VMAs that don't support it, such as I/O mappings
161 */
Ingo Molnarc12d2da2016-04-04 10:24:58 +0200162long get_user_pages(unsigned long start, unsigned long nr_pages,
Lorenzo Stoakes768ae302016-10-13 01:20:16 +0100163 unsigned int gup_flags, struct page **pages,
Michel Lespinasse28a35712013-02-22 16:35:55 -0800164 struct vm_area_struct **vmas)
Nick Pigginb291f002008-10-18 20:26:44 -0700165{
Lorenzo Stoakes768ae302016-10-13 01:20:16 +0100166 return __get_user_pages(current, current->mm, start, nr_pages,
167 gup_flags, pages, vmas, NULL);
Nick Pigginb291f002008-10-18 20:26:44 -0700168}
Ingo Molnarc12d2da2016-04-04 10:24:58 +0200169EXPORT_SYMBOL(get_user_pages);
Greg Ungerer66aa2b42005-09-12 11:18:10 +1000170
Ingo Molnarc12d2da2016-04-04 10:24:58 +0200171long get_user_pages_locked(unsigned long start, unsigned long nr_pages,
Lorenzo Stoakes3b913172016-10-13 01:20:14 +0100172 unsigned int gup_flags, struct page **pages,
Dave Hansencde70142016-02-12 13:01:55 -0800173 int *locked)
Andrea Arcangelif0818f42015-02-11 15:27:17 -0800174{
Lorenzo Stoakes768ae302016-10-13 01:20:16 +0100175 return get_user_pages(start, nr_pages, gup_flags, pages, NULL);
Andrea Arcangelif0818f42015-02-11 15:27:17 -0800176}
Ingo Molnarc12d2da2016-04-04 10:24:58 +0200177EXPORT_SYMBOL(get_user_pages_locked);
Andrea Arcangelif0818f42015-02-11 15:27:17 -0800178
Andrea Arcangeli0fd71a52015-02-11 15:27:20 -0800179long __get_user_pages_unlocked(struct task_struct *tsk, struct mm_struct *mm,
180 unsigned long start, unsigned long nr_pages,
Lorenzo Stoakesd4944b02016-10-13 01:20:12 +0100181 struct page **pages, unsigned int gup_flags)
Andrea Arcangelif0818f42015-02-11 15:27:17 -0800182{
183 long ret;
184 down_read(&mm->mmap_sem);
Dave Hansencde70142016-02-12 13:01:55 -0800185 ret = __get_user_pages(tsk, mm, start, nr_pages, gup_flags, pages,
186 NULL, NULL);
Andrea Arcangelif0818f42015-02-11 15:27:17 -0800187 up_read(&mm->mmap_sem);
188 return ret;
189}
Andrea Arcangeli0fd71a52015-02-11 15:27:20 -0800190EXPORT_SYMBOL(__get_user_pages_unlocked);
191
Ingo Molnarc12d2da2016-04-04 10:24:58 +0200192long get_user_pages_unlocked(unsigned long start, unsigned long nr_pages,
Lorenzo Stoakesc1641542016-10-13 01:20:13 +0100193 struct page **pages, unsigned int gup_flags)
Andrea Arcangeli0fd71a52015-02-11 15:27:20 -0800194{
Dave Hansencde70142016-02-12 13:01:55 -0800195 return __get_user_pages_unlocked(current, current->mm, start, nr_pages,
Lorenzo Stoakesc1641542016-10-13 01:20:13 +0100196 pages, gup_flags);
Andrea Arcangeli0fd71a52015-02-11 15:27:20 -0800197}
Ingo Molnarc12d2da2016-04-04 10:24:58 +0200198EXPORT_SYMBOL(get_user_pages_unlocked);
Andrea Arcangelif0818f42015-02-11 15:27:17 -0800199
Paul Mundtdfc2f912009-06-26 04:31:57 +0900200/**
201 * follow_pfn - look up PFN at a user virtual address
202 * @vma: memory mapping
203 * @address: user virtual address
204 * @pfn: location to store found PFN
205 *
206 * Only IO mappings and raw PFN mappings are allowed.
207 *
208 * Returns zero and the pfn at @pfn on success, -ve otherwise.
209 */
210int follow_pfn(struct vm_area_struct *vma, unsigned long address,
211 unsigned long *pfn)
212{
213 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
214 return -EINVAL;
215
216 *pfn = address >> PAGE_SHIFT;
217 return 0;
218}
219EXPORT_SYMBOL(follow_pfn);
220
Joonsoo Kimf1c40692013-04-29 15:07:37 -0700221LIST_HEAD(vmap_area_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800223void vfree(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224{
225 kfree(addr);
226}
Paul Mundtb5073172007-07-21 04:37:25 -0700227EXPORT_SYMBOL(vfree);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228
Al Virodd0fc662005-10-07 07:46:04 +0100229void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230{
231 /*
Robert P. J. Day85186092007-10-19 23:11:38 +0200232 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
233 * returns only a logical address.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 */
Nick Piggin84097512006-03-22 00:08:34 -0800235 return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236}
Paul Mundtb5073172007-07-21 04:37:25 -0700237EXPORT_SYMBOL(__vmalloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238
Paul Mundtf905bc42008-02-04 22:29:59 -0800239void *vmalloc_user(unsigned long size)
240{
241 void *ret;
242
243 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
244 PAGE_KERNEL);
245 if (ret) {
246 struct vm_area_struct *vma;
247
248 down_write(&current->mm->mmap_sem);
249 vma = find_vma(current->mm, (unsigned long)ret);
250 if (vma)
251 vma->vm_flags |= VM_USERMAP;
252 up_write(&current->mm->mmap_sem);
253 }
254
255 return ret;
256}
257EXPORT_SYMBOL(vmalloc_user);
258
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800259struct page *vmalloc_to_page(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260{
261 return virt_to_page(addr);
262}
Paul Mundtb5073172007-07-21 04:37:25 -0700263EXPORT_SYMBOL(vmalloc_to_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800265unsigned long vmalloc_to_pfn(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266{
267 return page_to_pfn(virt_to_page(addr));
268}
Paul Mundtb5073172007-07-21 04:37:25 -0700269EXPORT_SYMBOL(vmalloc_to_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270
271long vread(char *buf, char *addr, unsigned long count)
272{
Chen Gang9bde9162013-07-03 15:02:36 -0700273 /* Don't allow overflow */
274 if ((unsigned long) buf + count < count)
275 count = -(unsigned long) buf;
276
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 memcpy(buf, addr, count);
278 return count;
279}
280
281long vwrite(char *buf, char *addr, unsigned long count)
282{
283 /* Don't allow overflow */
284 if ((unsigned long) addr + count < count)
285 count = -(unsigned long) addr;
286
287 memcpy(addr, buf, count);
Choi Gi-yongac714902014-04-07 15:37:36 -0700288 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289}
290
291/*
Masahiro Yamadae1c05062015-07-07 10:14:59 +0900292 * vmalloc - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293 *
294 * @size: allocation size
295 *
296 * Allocate enough pages to cover @size from the page level
Masahiro Yamadae1c05062015-07-07 10:14:59 +0900297 * allocator and map them into contiguous kernel virtual space.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +0200299 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 * use __vmalloc() instead.
301 */
302void *vmalloc(unsigned long size)
303{
304 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
305}
Andrew Mortonf6138882006-02-28 16:59:18 -0800306EXPORT_SYMBOL(vmalloc);
307
Dave Younge1ca7782010-10-26 14:22:06 -0700308/*
Masahiro Yamadae1c05062015-07-07 10:14:59 +0900309 * vzalloc - allocate virtually contiguous memory with zero fill
Dave Younge1ca7782010-10-26 14:22:06 -0700310 *
311 * @size: allocation size
312 *
313 * Allocate enough pages to cover @size from the page level
Masahiro Yamadae1c05062015-07-07 10:14:59 +0900314 * allocator and map them into contiguous kernel virtual space.
Dave Younge1ca7782010-10-26 14:22:06 -0700315 * The memory allocated is set to zero.
316 *
317 * For tight control over page level allocator and protection flags
318 * use __vmalloc() instead.
319 */
320void *vzalloc(unsigned long size)
321{
322 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
323 PAGE_KERNEL);
324}
325EXPORT_SYMBOL(vzalloc);
326
327/**
328 * vmalloc_node - allocate memory on a specific node
329 * @size: allocation size
330 * @node: numa node
331 *
332 * Allocate enough pages to cover @size from the page level
333 * allocator and map them into contiguous kernel virtual space.
334 *
335 * For tight control over page level allocator and protection flags
336 * use __vmalloc() instead.
337 */
Andrew Mortonf6138882006-02-28 16:59:18 -0800338void *vmalloc_node(unsigned long size, int node)
339{
340 return vmalloc(size);
341}
Paul Mundt9a14f652010-12-24 11:50:34 +0900342EXPORT_SYMBOL(vmalloc_node);
Dave Younge1ca7782010-10-26 14:22:06 -0700343
344/**
345 * vzalloc_node - allocate memory on a specific node with zero fill
346 * @size: allocation size
347 * @node: numa node
348 *
349 * Allocate enough pages to cover @size from the page level
350 * allocator and map them into contiguous kernel virtual space.
351 * The memory allocated is set to zero.
352 *
353 * For tight control over page level allocator and protection flags
354 * use __vmalloc() instead.
355 */
356void *vzalloc_node(unsigned long size, int node)
357{
358 return vzalloc(size);
359}
360EXPORT_SYMBOL(vzalloc_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361
Paul Mundt1af446e2008-08-04 16:01:47 +0900362#ifndef PAGE_KERNEL_EXEC
363# define PAGE_KERNEL_EXEC PAGE_KERNEL
364#endif
365
366/**
367 * vmalloc_exec - allocate virtually contiguous, executable memory
368 * @size: allocation size
369 *
370 * Kernel-internal function to allocate enough pages to cover @size
371 * the page level allocator and map them into contiguous and
372 * executable kernel virtual space.
373 *
374 * For tight control over page level allocator and protection flags
375 * use __vmalloc() instead.
376 */
377
378void *vmalloc_exec(unsigned long size)
379{
380 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
381}
382
Paul Mundtb5073172007-07-21 04:37:25 -0700383/**
384 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 * @size: allocation size
386 *
387 * Allocate enough 32bit PA addressable pages to cover @size from the
Masahiro Yamadae1c05062015-07-07 10:14:59 +0900388 * page level allocator and map them into contiguous kernel virtual space.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 */
390void *vmalloc_32(unsigned long size)
391{
392 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
393}
Paul Mundtb5073172007-07-21 04:37:25 -0700394EXPORT_SYMBOL(vmalloc_32);
395
396/**
397 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
398 * @size: allocation size
399 *
400 * The resulting memory area is 32bit addressable and zeroed so it can be
401 * mapped to userspace without leaking data.
Paul Mundtf905bc42008-02-04 22:29:59 -0800402 *
403 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
404 * remap_vmalloc_range() are permissible.
Paul Mundtb5073172007-07-21 04:37:25 -0700405 */
406void *vmalloc_32_user(unsigned long size)
407{
Paul Mundtf905bc42008-02-04 22:29:59 -0800408 /*
409 * We'll have to sort out the ZONE_DMA bits for 64-bit,
410 * but for now this can simply use vmalloc_user() directly.
411 */
412 return vmalloc_user(size);
Paul Mundtb5073172007-07-21 04:37:25 -0700413}
414EXPORT_SYMBOL(vmalloc_32_user);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415
416void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
417{
418 BUG();
419 return NULL;
420}
Paul Mundtb5073172007-07-21 04:37:25 -0700421EXPORT_SYMBOL(vmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800423void vunmap(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424{
425 BUG();
426}
Paul Mundtb5073172007-07-21 04:37:25 -0700427EXPORT_SYMBOL(vunmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428
Paul Mundteb6434d2009-01-21 17:45:47 +0900429void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
430{
431 BUG();
432 return NULL;
433}
434EXPORT_SYMBOL(vm_map_ram);
435
436void vm_unmap_ram(const void *mem, unsigned int count)
437{
438 BUG();
439}
440EXPORT_SYMBOL(vm_unmap_ram);
441
442void vm_unmap_aliases(void)
443{
444}
445EXPORT_SYMBOL_GPL(vm_unmap_aliases);
446
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447/*
Joerg Roedel8c59bdc2020-03-21 18:22:41 -0700448 * Implement a stub for vmalloc_sync_[un]mapping() if the architecture
449 * chose not to have one.
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -0700450 */
Joerg Roedel8c59bdc2020-03-21 18:22:41 -0700451void __weak vmalloc_sync_mappings(void)
452{
453}
454
455void __weak vmalloc_sync_unmappings(void)
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -0700456{
457}
458
Paul Mundt29c185e2010-12-24 12:08:30 +0900459/**
460 * alloc_vm_area - allocate a range of kernel address space
461 * @size: size of the area
462 *
463 * Returns: NULL on failure, vm_struct on success
464 *
465 * This function reserves a range of kernel address space, and
466 * allocates pagetables to map that range. No actual mappings
467 * are created. If the kernel address space is not shared
468 * between processes, it syncs the pagetable across all
469 * processes.
470 */
David Vrabelcd129092011-09-29 16:53:32 +0100471struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
Paul Mundt29c185e2010-12-24 12:08:30 +0900472{
473 BUG();
474 return NULL;
475}
476EXPORT_SYMBOL_GPL(alloc_vm_area);
477
478void free_vm_area(struct vm_struct *area)
479{
480 BUG();
481}
482EXPORT_SYMBOL_GPL(free_vm_area);
483
Paul Mundtb5073172007-07-21 04:37:25 -0700484int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
485 struct page *page)
486{
487 return -EINVAL;
488}
489EXPORT_SYMBOL(vm_insert_page);
490
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -0700491/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 * sys_brk() for the most part doesn't need the global kernel
493 * lock, except when an application is doing something nasty
494 * like trying to un-brk an area that has already been mapped
495 * to a regular file. in this case, the unmapping will need
496 * to invoke file system routines that need the global lock.
497 */
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100498SYSCALL_DEFINE1(brk, unsigned long, brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499{
500 struct mm_struct *mm = current->mm;
501
502 if (brk < mm->start_brk || brk > mm->context.end_brk)
503 return mm->brk;
504
505 if (mm->brk == brk)
506 return mm->brk;
507
508 /*
509 * Always allow shrinking brk
510 */
511 if (brk <= mm->brk) {
512 mm->brk = brk;
513 return brk;
514 }
515
516 /*
517 * Ok, looks good - let it rip.
518 */
Mike Frysingercfe79c02010-01-06 17:23:23 +0000519 flush_icache_range(mm->brk, brk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 return mm->brk = brk;
521}
522
David Howells8feae132009-01-08 12:04:47 +0000523/*
524 * initialise the VMA and region record slabs
525 */
526void __init mmap_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527{
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -0700528 int ret;
529
Tejun Heo908c7f12014-09-08 09:51:29 +0900530 ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -0700531 VM_BUG_ON(ret);
Vladimir Davydov5d097052016-01-14 15:18:21 -0800532 vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC|SLAB_ACCOUNT);
David Howells8feae132009-01-08 12:04:47 +0000533}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534
David Howells8feae132009-01-08 12:04:47 +0000535/*
536 * validate the region tree
537 * - the caller must hold the region lock
538 */
539#ifdef CONFIG_DEBUG_NOMMU_REGIONS
540static noinline void validate_nommu_regions(void)
541{
542 struct vm_region *region, *last;
543 struct rb_node *p, *lastp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544
David Howells8feae132009-01-08 12:04:47 +0000545 lastp = rb_first(&nommu_region_tree);
546 if (!lastp)
547 return;
548
549 last = rb_entry(lastp, struct vm_region, vm_rb);
Geliang Tangc9427bc2015-11-05 18:48:38 -0800550 BUG_ON(last->vm_end <= last->vm_start);
551 BUG_ON(last->vm_top < last->vm_end);
David Howells8feae132009-01-08 12:04:47 +0000552
553 while ((p = rb_next(lastp))) {
554 region = rb_entry(p, struct vm_region, vm_rb);
555 last = rb_entry(lastp, struct vm_region, vm_rb);
556
Geliang Tangc9427bc2015-11-05 18:48:38 -0800557 BUG_ON(region->vm_end <= region->vm_start);
558 BUG_ON(region->vm_top < region->vm_end);
559 BUG_ON(region->vm_start < last->vm_top);
David Howells8feae132009-01-08 12:04:47 +0000560
561 lastp = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 }
563}
David Howells8feae132009-01-08 12:04:47 +0000564#else
David Howells33e5d7692009-04-02 16:56:32 -0700565static void validate_nommu_regions(void)
566{
567}
David Howells8feae132009-01-08 12:04:47 +0000568#endif
569
570/*
571 * add a region into the global tree
572 */
573static void add_nommu_region(struct vm_region *region)
574{
575 struct vm_region *pregion;
576 struct rb_node **p, *parent;
577
578 validate_nommu_regions();
579
David Howells8feae132009-01-08 12:04:47 +0000580 parent = NULL;
581 p = &nommu_region_tree.rb_node;
582 while (*p) {
583 parent = *p;
584 pregion = rb_entry(parent, struct vm_region, vm_rb);
585 if (region->vm_start < pregion->vm_start)
586 p = &(*p)->rb_left;
587 else if (region->vm_start > pregion->vm_start)
588 p = &(*p)->rb_right;
589 else if (pregion == region)
590 return;
591 else
592 BUG();
593 }
594
595 rb_link_node(&region->vm_rb, parent, p);
596 rb_insert_color(&region->vm_rb, &nommu_region_tree);
597
598 validate_nommu_regions();
599}
600
601/*
602 * delete a region from the global tree
603 */
604static void delete_nommu_region(struct vm_region *region)
605{
606 BUG_ON(!nommu_region_tree.rb_node);
607
608 validate_nommu_regions();
609 rb_erase(&region->vm_rb, &nommu_region_tree);
610 validate_nommu_regions();
611}
612
613/*
614 * free a contiguous series of pages
615 */
616static void free_page_series(unsigned long from, unsigned long to)
617{
618 for (; from < to; from += PAGE_SIZE) {
619 struct page *page = virt_to_page(from);
620
David Howells33e5d7692009-04-02 16:56:32 -0700621 atomic_long_dec(&mmap_pages_allocated);
David Howells8feae132009-01-08 12:04:47 +0000622 put_page(page);
623 }
624}
625
626/*
627 * release a reference to a region
David Howells33e5d7692009-04-02 16:56:32 -0700628 * - the caller must hold the region semaphore for writing, which this releases
Paul Mundtdd8632a2009-01-08 12:04:47 +0000629 * - the region may not have been added to the tree yet, in which case vm_top
David Howells8feae132009-01-08 12:04:47 +0000630 * will equal vm_start
631 */
632static void __put_nommu_region(struct vm_region *region)
633 __releases(nommu_region_sem)
634{
David Howells8feae132009-01-08 12:04:47 +0000635 BUG_ON(!nommu_region_tree.rb_node);
636
David Howells1e2ae592010-01-15 17:01:33 -0800637 if (--region->vm_usage == 0) {
Paul Mundtdd8632a2009-01-08 12:04:47 +0000638 if (region->vm_top > region->vm_start)
David Howells8feae132009-01-08 12:04:47 +0000639 delete_nommu_region(region);
640 up_write(&nommu_region_sem);
641
642 if (region->vm_file)
643 fput(region->vm_file);
644
645 /* IO memory and memory shared directly out of the pagecache
646 * from ramfs/tmpfs mustn't be released here */
Leon Romanovsky22cc8772015-06-24 16:57:47 -0700647 if (region->vm_flags & VM_MAPPED_COPY)
Paul Mundtdd8632a2009-01-08 12:04:47 +0000648 free_page_series(region->vm_start, region->vm_top);
David Howells8feae132009-01-08 12:04:47 +0000649 kmem_cache_free(vm_region_jar, region);
650 } else {
651 up_write(&nommu_region_sem);
652 }
653}
654
655/*
656 * release a reference to a region
657 */
658static void put_nommu_region(struct vm_region *region)
659{
660 down_write(&nommu_region_sem);
661 __put_nommu_region(region);
662}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663
David Howells30340972006-09-27 01:50:20 -0700664/*
Bernd Schmidteb8cdec2009-09-21 17:03:57 -0700665 * update protection on a vma
666 */
667static void protect_vma(struct vm_area_struct *vma, unsigned long flags)
668{
669#ifdef CONFIG_MPU
670 struct mm_struct *mm = vma->vm_mm;
671 long start = vma->vm_start & PAGE_MASK;
672 while (start < vma->vm_end) {
673 protect_page(mm, start, flags);
674 start += PAGE_SIZE;
675 }
676 update_protections(mm);
677#endif
678}
679
680/*
David Howells30340972006-09-27 01:50:20 -0700681 * add a VMA into a process's mm_struct in the appropriate place in the list
David Howells8feae132009-01-08 12:04:47 +0000682 * and tree and add to the address space's page tree also if not an anonymous
683 * page
David Howells30340972006-09-27 01:50:20 -0700684 * - should be called with mm->mmap_sem held writelocked
685 */
David Howells8feae132009-01-08 12:04:47 +0000686static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
David Howells30340972006-09-27 01:50:20 -0700687{
Namhyung Kim6038def2011-05-24 17:11:22 -0700688 struct vm_area_struct *pvma, *prev;
David Howells8feae132009-01-08 12:04:47 +0000689 struct address_space *mapping;
Namhyung Kim6038def2011-05-24 17:11:22 -0700690 struct rb_node **p, *parent, *rb_prev;
David Howells30340972006-09-27 01:50:20 -0700691
David Howells8feae132009-01-08 12:04:47 +0000692 BUG_ON(!vma->vm_region);
693
694 mm->map_count++;
695 vma->vm_mm = mm;
696
Bernd Schmidteb8cdec2009-09-21 17:03:57 -0700697 protect_vma(vma, vma->vm_flags);
698
David Howells8feae132009-01-08 12:04:47 +0000699 /* add the VMA to the mapping */
700 if (vma->vm_file) {
701 mapping = vma->vm_file->f_mapping;
702
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800703 i_mmap_lock_write(mapping);
David Howells8feae132009-01-08 12:04:47 +0000704 flush_dcache_mmap_lock(mapping);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700705 vma_interval_tree_insert(vma, &mapping->i_mmap);
David Howells8feae132009-01-08 12:04:47 +0000706 flush_dcache_mmap_unlock(mapping);
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800707 i_mmap_unlock_write(mapping);
David Howells8feae132009-01-08 12:04:47 +0000708 }
709
710 /* add the VMA to the tree */
Namhyung Kim6038def2011-05-24 17:11:22 -0700711 parent = rb_prev = NULL;
David Howells8feae132009-01-08 12:04:47 +0000712 p = &mm->mm_rb.rb_node;
713 while (*p) {
714 parent = *p;
715 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
716
717 /* sort by: start addr, end addr, VMA struct addr in that order
718 * (the latter is necessary as we may get identical VMAs) */
719 if (vma->vm_start < pvma->vm_start)
720 p = &(*p)->rb_left;
Namhyung Kim6038def2011-05-24 17:11:22 -0700721 else if (vma->vm_start > pvma->vm_start) {
722 rb_prev = parent;
David Howells8feae132009-01-08 12:04:47 +0000723 p = &(*p)->rb_right;
Namhyung Kim6038def2011-05-24 17:11:22 -0700724 } else if (vma->vm_end < pvma->vm_end)
David Howells8feae132009-01-08 12:04:47 +0000725 p = &(*p)->rb_left;
Namhyung Kim6038def2011-05-24 17:11:22 -0700726 else if (vma->vm_end > pvma->vm_end) {
727 rb_prev = parent;
David Howells8feae132009-01-08 12:04:47 +0000728 p = &(*p)->rb_right;
Namhyung Kim6038def2011-05-24 17:11:22 -0700729 } else if (vma < pvma)
David Howells8feae132009-01-08 12:04:47 +0000730 p = &(*p)->rb_left;
Namhyung Kim6038def2011-05-24 17:11:22 -0700731 else if (vma > pvma) {
732 rb_prev = parent;
David Howells8feae132009-01-08 12:04:47 +0000733 p = &(*p)->rb_right;
Namhyung Kim6038def2011-05-24 17:11:22 -0700734 } else
David Howells8feae132009-01-08 12:04:47 +0000735 BUG();
736 }
737
738 rb_link_node(&vma->vm_rb, parent, p);
739 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
740
741 /* add VMA to the VMA list also */
Namhyung Kim6038def2011-05-24 17:11:22 -0700742 prev = NULL;
743 if (rb_prev)
744 prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
David Howells30340972006-09-27 01:50:20 -0700745
Namhyung Kim6038def2011-05-24 17:11:22 -0700746 __vma_link_list(mm, vma, prev, parent);
David Howells8feae132009-01-08 12:04:47 +0000747}
748
749/*
750 * delete a VMA from its owning mm_struct and address space
751 */
752static void delete_vma_from_mm(struct vm_area_struct *vma)
753{
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700754 int i;
David Howells8feae132009-01-08 12:04:47 +0000755 struct address_space *mapping;
756 struct mm_struct *mm = vma->vm_mm;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700757 struct task_struct *curr = current;
David Howells8feae132009-01-08 12:04:47 +0000758
Bernd Schmidteb8cdec2009-09-21 17:03:57 -0700759 protect_vma(vma, 0);
760
David Howells8feae132009-01-08 12:04:47 +0000761 mm->map_count--;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700762 for (i = 0; i < VMACACHE_SIZE; i++) {
763 /* if the vma is cached, invalidate the entire cache */
764 if (curr->vmacache[i] == vma) {
Steven Miaoe020d5b2014-06-23 13:22:02 -0700765 vmacache_invalidate(mm);
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700766 break;
767 }
768 }
David Howells8feae132009-01-08 12:04:47 +0000769
770 /* remove the VMA from the mapping */
771 if (vma->vm_file) {
772 mapping = vma->vm_file->f_mapping;
773
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800774 i_mmap_lock_write(mapping);
David Howells8feae132009-01-08 12:04:47 +0000775 flush_dcache_mmap_lock(mapping);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700776 vma_interval_tree_remove(vma, &mapping->i_mmap);
David Howells8feae132009-01-08 12:04:47 +0000777 flush_dcache_mmap_unlock(mapping);
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800778 i_mmap_unlock_write(mapping);
David Howells8feae132009-01-08 12:04:47 +0000779 }
780
781 /* remove from the MM's tree and list */
782 rb_erase(&vma->vm_rb, &mm->mm_rb);
Namhyung Kimb951bf22011-05-24 17:11:23 -0700783
784 if (vma->vm_prev)
785 vma->vm_prev->vm_next = vma->vm_next;
786 else
787 mm->mmap = vma->vm_next;
788
789 if (vma->vm_next)
790 vma->vm_next->vm_prev = vma->vm_prev;
David Howells8feae132009-01-08 12:04:47 +0000791}
792
793/*
794 * destroy a VMA record
795 */
796static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
797{
David Howells8feae132009-01-08 12:04:47 +0000798 if (vma->vm_ops && vma->vm_ops->close)
799 vma->vm_ops->close(vma);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -0700800 if (vma->vm_file)
David Howells8feae132009-01-08 12:04:47 +0000801 fput(vma->vm_file);
David Howells8feae132009-01-08 12:04:47 +0000802 put_nommu_region(vma->vm_region);
803 kmem_cache_free(vm_area_cachep, vma);
David Howells30340972006-09-27 01:50:20 -0700804}
805
806/*
807 * look up the first VMA in which addr resides, NULL if none
808 * - should be called with mm->mmap_sem at least held readlocked
809 */
810struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
811{
David Howells8feae132009-01-08 12:04:47 +0000812 struct vm_area_struct *vma;
David Howells30340972006-09-27 01:50:20 -0700813
David Howells8feae132009-01-08 12:04:47 +0000814 /* check the cache first */
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700815 vma = vmacache_find(mm, addr);
816 if (likely(vma))
David Howells8feae132009-01-08 12:04:47 +0000817 return vma;
818
Namhyung Kime922c4c2011-05-24 17:11:24 -0700819 /* trawl the list (there may be multiple mappings in which addr
David Howells8feae132009-01-08 12:04:47 +0000820 * resides) */
Namhyung Kime922c4c2011-05-24 17:11:24 -0700821 for (vma = mm->mmap; vma; vma = vma->vm_next) {
David Howells8feae132009-01-08 12:04:47 +0000822 if (vma->vm_start > addr)
823 return NULL;
824 if (vma->vm_end > addr) {
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700825 vmacache_update(addr, vma);
David Howells8feae132009-01-08 12:04:47 +0000826 return vma;
827 }
David Howells30340972006-09-27 01:50:20 -0700828 }
829
David Howells30340972006-09-27 01:50:20 -0700830 return NULL;
831}
832EXPORT_SYMBOL(find_vma);
833
834/*
David Howells930e6522006-09-27 01:50:22 -0700835 * find a VMA
836 * - we don't extend stack VMAs under NOMMU conditions
837 */
838struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
839{
David Howells7561e8c2010-03-25 16:48:38 +0000840 return find_vma(mm, addr);
David Howells930e6522006-09-27 01:50:22 -0700841}
842
David Howells8feae132009-01-08 12:04:47 +0000843/*
844 * expand a stack to a given address
845 * - not supported under NOMMU conditions
846 */
Greg Ungerer57c8f632007-07-15 23:38:28 -0700847int expand_stack(struct vm_area_struct *vma, unsigned long address)
848{
849 return -ENOMEM;
850}
851
David Howells930e6522006-09-27 01:50:22 -0700852/*
David Howells6fa5f802006-09-27 01:50:21 -0700853 * look up the first VMA exactly that exactly matches addr
854 * - should be called with mm->mmap_sem at least held readlocked
855 */
David Howells8feae132009-01-08 12:04:47 +0000856static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
857 unsigned long addr,
858 unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859{
860 struct vm_area_struct *vma;
David Howells8feae132009-01-08 12:04:47 +0000861 unsigned long end = addr + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862
David Howells8feae132009-01-08 12:04:47 +0000863 /* check the cache first */
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700864 vma = vmacache_find_exact(mm, addr, end);
865 if (vma)
David Howells8feae132009-01-08 12:04:47 +0000866 return vma;
867
Namhyung Kime922c4c2011-05-24 17:11:24 -0700868 /* trawl the list (there may be multiple mappings in which addr
David Howells8feae132009-01-08 12:04:47 +0000869 * resides) */
Namhyung Kime922c4c2011-05-24 17:11:24 -0700870 for (vma = mm->mmap; vma; vma = vma->vm_next) {
David Howells8feae132009-01-08 12:04:47 +0000871 if (vma->vm_start < addr)
872 continue;
873 if (vma->vm_start > addr)
874 return NULL;
875 if (vma->vm_end == end) {
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700876 vmacache_update(addr, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 return vma;
David Howells8feae132009-01-08 12:04:47 +0000878 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 }
880
881 return NULL;
882}
883
David Howells30340972006-09-27 01:50:20 -0700884/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 * determine whether a mapping should be permitted and, if so, what sort of
886 * mapping we're capable of supporting
887 */
888static int validate_mmap_request(struct file *file,
889 unsigned long addr,
890 unsigned long len,
891 unsigned long prot,
892 unsigned long flags,
893 unsigned long pgoff,
894 unsigned long *_capabilities)
895{
David Howells8feae132009-01-08 12:04:47 +0000896 unsigned long capabilities, rlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 int ret;
898
899 /* do the simple checks first */
Leon Romanovsky22cc8772015-06-24 16:57:47 -0700900 if (flags & MAP_FIXED)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902
903 if ((flags & MAP_TYPE) != MAP_PRIVATE &&
904 (flags & MAP_TYPE) != MAP_SHARED)
905 return -EINVAL;
906
Mike Frysingerf81cff02006-12-06 12:02:59 +1000907 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 return -EINVAL;
909
Mike Frysingerf81cff02006-12-06 12:02:59 +1000910 /* Careful about overflows.. */
David Howells8feae132009-01-08 12:04:47 +0000911 rlen = PAGE_ALIGN(len);
912 if (!rlen || rlen > TASK_SIZE)
Mike Frysingerf81cff02006-12-06 12:02:59 +1000913 return -ENOMEM;
914
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 /* offset overflow? */
David Howells8feae132009-01-08 12:04:47 +0000916 if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
Mike Frysingerf81cff02006-12-06 12:02:59 +1000917 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918
919 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 /* files must support mmap */
Al Viro72c2d532013-09-22 16:27:52 -0400921 if (!file->f_op->mmap)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 return -ENODEV;
923
924 /* work out if what we've got could possibly be shared
925 * - we support chardevs that provide their own "memory"
926 * - we support files/blockdevs that are memory backed
927 */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100928 if (file->f_op->mmap_capabilities) {
929 capabilities = file->f_op->mmap_capabilities(file);
930 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 /* no explicit capabilities set, so assume some
932 * defaults */
Al Viro496ad9a2013-01-23 17:07:38 -0500933 switch (file_inode(file)->i_mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 case S_IFREG:
935 case S_IFBLK:
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100936 capabilities = NOMMU_MAP_COPY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 break;
938
939 case S_IFCHR:
940 capabilities =
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100941 NOMMU_MAP_DIRECT |
942 NOMMU_MAP_READ |
943 NOMMU_MAP_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 break;
945
946 default:
947 return -EINVAL;
948 }
949 }
950
951 /* eliminate any capabilities that we can't support on this
952 * device */
953 if (!file->f_op->get_unmapped_area)
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100954 capabilities &= ~NOMMU_MAP_DIRECT;
Al Viro6e242a12015-03-31 12:35:13 -0400955 if (!(file->f_mode & FMODE_CAN_READ))
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100956 capabilities &= ~NOMMU_MAP_COPY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957
Graff Yang28d7a6a2009-08-18 14:11:17 -0700958 /* The file shall have been opened with read permission. */
959 if (!(file->f_mode & FMODE_READ))
960 return -EACCES;
961
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 if (flags & MAP_SHARED) {
963 /* do checks for writing, appending and locking */
964 if ((prot & PROT_WRITE) &&
965 !(file->f_mode & FMODE_WRITE))
966 return -EACCES;
967
Al Viro496ad9a2013-01-23 17:07:38 -0500968 if (IS_APPEND(file_inode(file)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 (file->f_mode & FMODE_WRITE))
970 return -EACCES;
971
Jeff Laytond7a06982014-03-10 09:54:15 -0400972 if (locks_verify_locked(file))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973 return -EAGAIN;
974
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100975 if (!(capabilities & NOMMU_MAP_DIRECT))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976 return -ENODEV;
977
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 /* we mustn't privatise shared mappings */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100979 capabilities &= ~NOMMU_MAP_COPY;
Choi Gi-yongac714902014-04-07 15:37:36 -0700980 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 /* we're going to read the file into private memory we
982 * allocate */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100983 if (!(capabilities & NOMMU_MAP_COPY))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 return -ENODEV;
985
986 /* we don't permit a private writable mapping to be
987 * shared with the backing device */
988 if (prot & PROT_WRITE)
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100989 capabilities &= ~NOMMU_MAP_DIRECT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 }
991
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100992 if (capabilities & NOMMU_MAP_DIRECT) {
993 if (((prot & PROT_READ) && !(capabilities & NOMMU_MAP_READ)) ||
994 ((prot & PROT_WRITE) && !(capabilities & NOMMU_MAP_WRITE)) ||
995 ((prot & PROT_EXEC) && !(capabilities & NOMMU_MAP_EXEC))
Bernd Schmidt3c7b2042010-05-25 23:43:00 -0700996 ) {
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100997 capabilities &= ~NOMMU_MAP_DIRECT;
Bernd Schmidt3c7b2042010-05-25 23:43:00 -0700998 if (flags & MAP_SHARED) {
Leon Romanovsky22cc8772015-06-24 16:57:47 -0700999 pr_warn("MAP_SHARED not completely supported on !MMU\n");
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001000 return -EINVAL;
1001 }
1002 }
1003 }
1004
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 /* handle executable mappings and implied executable
1006 * mappings */
Eric W. Biederman90f85722015-06-29 14:42:03 -05001007 if (path_noexec(&file->f_path)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 if (prot & PROT_EXEC)
1009 return -EPERM;
Choi Gi-yongac714902014-04-07 15:37:36 -07001010 } else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 /* handle implication of PROT_EXEC by PROT_READ */
1012 if (current->personality & READ_IMPLIES_EXEC) {
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001013 if (capabilities & NOMMU_MAP_EXEC)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 prot |= PROT_EXEC;
1015 }
Choi Gi-yongac714902014-04-07 15:37:36 -07001016 } else if ((prot & PROT_READ) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 (prot & PROT_EXEC) &&
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001018 !(capabilities & NOMMU_MAP_EXEC)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 ) {
1020 /* backing file is not executable, try to copy */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001021 capabilities &= ~NOMMU_MAP_DIRECT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 }
Choi Gi-yongac714902014-04-07 15:37:36 -07001023 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 /* anonymous mappings are always memory backed and can be
1025 * privately mapped
1026 */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001027 capabilities = NOMMU_MAP_COPY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028
1029 /* handle PROT_EXEC implication by PROT_READ */
1030 if ((prot & PROT_READ) &&
1031 (current->personality & READ_IMPLIES_EXEC))
1032 prot |= PROT_EXEC;
1033 }
1034
1035 /* allow the security API to have its say */
Al Viroe5467852012-05-30 13:30:51 -04001036 ret = security_mmap_addr(addr);
1037 if (ret < 0)
1038 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039
1040 /* looks okay */
1041 *_capabilities = capabilities;
1042 return 0;
1043}
1044
1045/*
1046 * we've determined that we can make the mapping, now translate what we
1047 * now know into VMA flags
1048 */
1049static unsigned long determine_vm_flags(struct file *file,
1050 unsigned long prot,
1051 unsigned long flags,
1052 unsigned long capabilities)
1053{
1054 unsigned long vm_flags;
1055
Dave Hansene6bfb702016-02-12 13:02:31 -08001056 vm_flags = calc_vm_prot_bits(prot, 0) | calc_vm_flag_bits(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 /* vm_flags |= mm->def_flags; */
1058
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001059 if (!(capabilities & NOMMU_MAP_DIRECT)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 /* attempt to share read-only copies of mapped file chunks */
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001061 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062 if (file && !(prot & PROT_WRITE))
1063 vm_flags |= VM_MAYSHARE;
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001064 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 /* overlay a shareable mapping on the backing device or inode
1066 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1067 * romfs/cramfs */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001068 vm_flags |= VM_MAYSHARE | (capabilities & NOMMU_VMFLAGS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069 if (flags & MAP_SHARED)
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001070 vm_flags |= VM_SHARED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071 }
1072
1073 /* refuse to let anyone share private mappings with this process if
1074 * it's being traced - otherwise breakpoints set in it may interfere
1075 * with another untraced process
1076 */
Tejun Heoa288eec2011-06-17 16:50:37 +02001077 if ((flags & MAP_PRIVATE) && current->ptrace)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 vm_flags &= ~VM_MAYSHARE;
1079
1080 return vm_flags;
1081}
1082
1083/*
David Howells8feae132009-01-08 12:04:47 +00001084 * set up a shared mapping on a file (the driver or filesystem provides and
1085 * pins the storage)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 */
David Howells8feae132009-01-08 12:04:47 +00001087static int do_mmap_shared_file(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088{
1089 int ret;
1090
1091 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001092 if (ret == 0) {
1093 vma->vm_region->vm_top = vma->vm_region->vm_end;
David Howells645d83c2009-09-24 15:13:10 +01001094 return 0;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001095 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 if (ret != -ENOSYS)
1097 return ret;
1098
David Howells3fa30462010-03-23 13:35:21 -07001099 /* getting -ENOSYS indicates that direct mmap isn't possible (as
1100 * opposed to tried but failed) so we can only give a suitable error as
1101 * it's not possible to make a private copy if MAP_SHARED was given */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 return -ENODEV;
1103}
1104
1105/*
1106 * set up a private mapping or an anonymous shared mapping
1107 */
David Howells8feae132009-01-08 12:04:47 +00001108static int do_mmap_private(struct vm_area_struct *vma,
1109 struct vm_region *region,
David Howells645d83c2009-09-24 15:13:10 +01001110 unsigned long len,
1111 unsigned long capabilities)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112{
Joonsoo Kimdbc83582014-12-12 16:55:55 -08001113 unsigned long total, point;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114 void *base;
David Howells8feae132009-01-08 12:04:47 +00001115 int ret, order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116
1117 /* invoke the file's mapping function so that it can keep track of
1118 * shared mappings on devices or memory
1119 * - VM_MAYSHARE will be set if it may attempt to share
1120 */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001121 if (capabilities & NOMMU_MAP_DIRECT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001123 if (ret == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 /* shouldn't return success if we're not sharing */
Paul Mundtdd8632a2009-01-08 12:04:47 +00001125 BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1126 vma->vm_region->vm_top = vma->vm_region->vm_end;
David Howells645d83c2009-09-24 15:13:10 +01001127 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 }
Paul Mundtdd8632a2009-01-08 12:04:47 +00001129 if (ret != -ENOSYS)
1130 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131
1132 /* getting an ENOSYS error indicates that direct mmap isn't
1133 * possible (as opposed to tried but failed) so we'll try to
1134 * make a private copy of the data and map that instead */
1135 }
1136
David Howells8feae132009-01-08 12:04:47 +00001137
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138 /* allocate some memory to hold the mapping
1139 * - note that this may not return a page-aligned address if the object
1140 * we're allocating is smaller than a page
1141 */
Bob Liuf67d9b12011-05-24 17:12:56 -07001142 order = get_order(len);
David Howells8feae132009-01-08 12:04:47 +00001143 total = 1 << order;
Bob Liuf67d9b12011-05-24 17:12:56 -07001144 point = len >> PAGE_SHIFT;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001145
Joonsoo Kimdbc83582014-12-12 16:55:55 -08001146 /* we don't want to allocate a power-of-2 sized page set */
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001147 if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages)
Joonsoo Kimdbc83582014-12-12 16:55:55 -08001148 total = point;
David Howells8feae132009-01-08 12:04:47 +00001149
Joonsoo Kimda616532015-02-27 15:51:43 -08001150 base = alloc_pages_exact(total << PAGE_SHIFT, GFP_KERNEL);
Joonsoo Kimdbc83582014-12-12 16:55:55 -08001151 if (!base)
1152 goto enomem;
David Howells8feae132009-01-08 12:04:47 +00001153
Joonsoo Kimdbc83582014-12-12 16:55:55 -08001154 atomic_long_add(total, &mmap_pages_allocated);
1155
David Howells8feae132009-01-08 12:04:47 +00001156 region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1157 region->vm_start = (unsigned long) base;
Bob Liuf67d9b12011-05-24 17:12:56 -07001158 region->vm_end = region->vm_start + len;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001159 region->vm_top = region->vm_start + (total << PAGE_SHIFT);
David Howells8feae132009-01-08 12:04:47 +00001160
1161 vma->vm_start = region->vm_start;
1162 vma->vm_end = region->vm_start + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163
1164 if (vma->vm_file) {
1165 /* read the contents of a file into the copy */
1166 mm_segment_t old_fs;
1167 loff_t fpos;
1168
1169 fpos = vma->vm_pgoff;
1170 fpos <<= PAGE_SHIFT;
1171
1172 old_fs = get_fs();
1173 set_fs(KERNEL_DS);
Al Viro6e242a12015-03-31 12:35:13 -04001174 ret = __vfs_read(vma->vm_file, base, len, &fpos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 set_fs(old_fs);
1176
1177 if (ret < 0)
1178 goto error_free;
1179
1180 /* clear the last little bit */
Bob Liuf67d9b12011-05-24 17:12:56 -07001181 if (ret < len)
1182 memset(base + ret, 0, len - ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 }
1185
1186 return 0;
1187
1188error_free:
Namhyung Kim7223bb42011-05-24 17:11:26 -07001189 free_page_series(region->vm_start, region->vm_top);
David Howells8feae132009-01-08 12:04:47 +00001190 region->vm_start = vma->vm_start = 0;
1191 region->vm_end = vma->vm_end = 0;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001192 region->vm_top = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193 return ret;
1194
1195enomem:
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -07001196 pr_err("Allocation of length %lu from process %d (%s) failed\n",
Greg Ungerer05ae6fa2009-01-13 17:30:22 +10001197 len, current->pid, current->comm);
David Rientjes7bf02ea2011-05-24 17:11:16 -07001198 show_free_areas(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 return -ENOMEM;
1200}
1201
1202/*
1203 * handle mapping creation for uClinux
1204 */
Oleg Nesterov1fcfd8d2015-09-09 15:39:29 -07001205unsigned long do_mmap(struct file *file,
1206 unsigned long addr,
1207 unsigned long len,
1208 unsigned long prot,
1209 unsigned long flags,
1210 vm_flags_t vm_flags,
1211 unsigned long pgoff,
1212 unsigned long *populate)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213{
David Howells8feae132009-01-08 12:04:47 +00001214 struct vm_area_struct *vma;
1215 struct vm_region *region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 struct rb_node *rb;
Oleg Nesterov1fcfd8d2015-09-09 15:39:29 -07001217 unsigned long capabilities, result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218 int ret;
1219
Michel Lespinasse41badc12013-02-22 16:32:47 -08001220 *populate = 0;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001221
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 /* decide whether we should attempt the mapping, and if so what sort of
1223 * mapping */
1224 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1225 &capabilities);
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001226 if (ret < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 return ret;
1228
David Howells06aab5a2009-09-24 12:33:48 +01001229 /* we ignore the address hint */
1230 addr = 0;
Bob Liuf67d9b12011-05-24 17:12:56 -07001231 len = PAGE_ALIGN(len);
David Howells06aab5a2009-09-24 12:33:48 +01001232
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233 /* we've determined that we can make the mapping, now translate what we
1234 * now know into VMA flags */
Oleg Nesterov1fcfd8d2015-09-09 15:39:29 -07001235 vm_flags |= determine_vm_flags(file, prot, flags, capabilities);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236
David Howells8feae132009-01-08 12:04:47 +00001237 /* we're going to need to record the mapping */
1238 region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1239 if (!region)
1240 goto error_getting_region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241
David Howells8feae132009-01-08 12:04:47 +00001242 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1243 if (!vma)
1244 goto error_getting_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245
David Howells1e2ae592010-01-15 17:01:33 -08001246 region->vm_usage = 1;
David Howells8feae132009-01-08 12:04:47 +00001247 region->vm_flags = vm_flags;
1248 region->vm_pgoff = pgoff;
1249
Rik van Riel5beb4932010-03-05 13:42:07 -08001250 INIT_LIST_HEAD(&vma->anon_vma_chain);
David Howells8feae132009-01-08 12:04:47 +00001251 vma->vm_flags = vm_flags;
1252 vma->vm_pgoff = pgoff;
1253
1254 if (file) {
Al Virocb0942b2012-08-27 14:48:26 -04001255 region->vm_file = get_file(file);
1256 vma->vm_file = get_file(file);
David Howells8feae132009-01-08 12:04:47 +00001257 }
1258
1259 down_write(&nommu_region_sem);
1260
1261 /* if we want to share, we need to check for regions created by other
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 * mmap() calls that overlap with our proposed mapping
David Howells8feae132009-01-08 12:04:47 +00001263 * - we can only share with a superset match on most regular files
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 * - shared mappings on character devices and memory backed files are
1265 * permitted to overlap inexactly as far as we are concerned for in
1266 * these cases, sharing is handled in the driver or filesystem rather
1267 * than here
1268 */
1269 if (vm_flags & VM_MAYSHARE) {
David Howells8feae132009-01-08 12:04:47 +00001270 struct vm_region *pregion;
1271 unsigned long pglen, rpglen, pgend, rpgend, start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272
David Howells8feae132009-01-08 12:04:47 +00001273 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1274 pgend = pgoff + pglen;
David Howells165b2392007-03-22 00:11:24 -08001275
David Howells8feae132009-01-08 12:04:47 +00001276 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1277 pregion = rb_entry(rb, struct vm_region, vm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278
David Howells8feae132009-01-08 12:04:47 +00001279 if (!(pregion->vm_flags & VM_MAYSHARE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 continue;
1281
1282 /* search for overlapping mappings on the same file */
Al Viro496ad9a2013-01-23 17:07:38 -05001283 if (file_inode(pregion->vm_file) !=
1284 file_inode(file))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 continue;
1286
David Howells8feae132009-01-08 12:04:47 +00001287 if (pregion->vm_pgoff >= pgend)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288 continue;
1289
David Howells8feae132009-01-08 12:04:47 +00001290 rpglen = pregion->vm_end - pregion->vm_start;
1291 rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1292 rpgend = pregion->vm_pgoff + rpglen;
1293 if (pgoff >= rpgend)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294 continue;
1295
David Howells8feae132009-01-08 12:04:47 +00001296 /* handle inexactly overlapping matches between
1297 * mappings */
1298 if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1299 !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1300 /* new mapping is not a subset of the region */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001301 if (!(capabilities & NOMMU_MAP_DIRECT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302 goto sharing_violation;
1303 continue;
1304 }
1305
David Howells8feae132009-01-08 12:04:47 +00001306 /* we've found a region we can share */
David Howells1e2ae592010-01-15 17:01:33 -08001307 pregion->vm_usage++;
David Howells8feae132009-01-08 12:04:47 +00001308 vma->vm_region = pregion;
1309 start = pregion->vm_start;
1310 start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1311 vma->vm_start = start;
1312 vma->vm_end = start + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001314 if (pregion->vm_flags & VM_MAPPED_COPY)
David Howells8feae132009-01-08 12:04:47 +00001315 vma->vm_flags |= VM_MAPPED_COPY;
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001316 else {
David Howells8feae132009-01-08 12:04:47 +00001317 ret = do_mmap_shared_file(vma);
1318 if (ret < 0) {
1319 vma->vm_region = NULL;
1320 vma->vm_start = 0;
1321 vma->vm_end = 0;
David Howells1e2ae592010-01-15 17:01:33 -08001322 pregion->vm_usage--;
David Howells8feae132009-01-08 12:04:47 +00001323 pregion = NULL;
1324 goto error_just_free;
1325 }
1326 }
1327 fput(region->vm_file);
1328 kmem_cache_free(vm_region_jar, region);
1329 region = pregion;
1330 result = start;
1331 goto share;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 }
1333
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334 /* obtain the address at which to make a shared mapping
1335 * - this is the hook for quasi-memory character devices to
1336 * tell us the location of a shared mapping
1337 */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001338 if (capabilities & NOMMU_MAP_DIRECT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 addr = file->f_op->get_unmapped_area(file, addr, len,
1340 pgoff, flags);
Namhyung Kimbb005a52011-05-24 17:11:27 -07001341 if (IS_ERR_VALUE(addr)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 ret = addr;
Namhyung Kimbb005a52011-05-24 17:11:27 -07001343 if (ret != -ENOSYS)
David Howells8feae132009-01-08 12:04:47 +00001344 goto error_just_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345
1346 /* the driver refused to tell us where to site
1347 * the mapping so we'll have to attempt to copy
1348 * it */
Namhyung Kimbb005a52011-05-24 17:11:27 -07001349 ret = -ENODEV;
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001350 if (!(capabilities & NOMMU_MAP_COPY))
David Howells8feae132009-01-08 12:04:47 +00001351 goto error_just_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001353 capabilities &= ~NOMMU_MAP_DIRECT;
David Howells8feae132009-01-08 12:04:47 +00001354 } else {
1355 vma->vm_start = region->vm_start = addr;
1356 vma->vm_end = region->vm_end = addr + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 }
1358 }
1359 }
1360
David Howells8feae132009-01-08 12:04:47 +00001361 vma->vm_region = region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362
David Howells645d83c2009-09-24 15:13:10 +01001363 /* set up the mapping
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001364 * - the region is filled in if NOMMU_MAP_DIRECT is still set
David Howells645d83c2009-09-24 15:13:10 +01001365 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 if (file && vma->vm_flags & VM_SHARED)
David Howells8feae132009-01-08 12:04:47 +00001367 ret = do_mmap_shared_file(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368 else
David Howells645d83c2009-09-24 15:13:10 +01001369 ret = do_mmap_private(vma, region, len, capabilities);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 if (ret < 0)
David Howells645d83c2009-09-24 15:13:10 +01001371 goto error_just_free;
1372 add_nommu_region(region);
David Howells8feae132009-01-08 12:04:47 +00001373
Jie Zhangea637632009-12-14 18:00:02 -08001374 /* clear anonymous mappings that don't ask for uninitialized data */
1375 if (!vma->vm_file && !(flags & MAP_UNINITIALIZED))
1376 memset((void *)region->vm_start, 0,
1377 region->vm_end - region->vm_start);
1378
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 /* okay... we have a mapping; now we have to register it */
David Howells8feae132009-01-08 12:04:47 +00001380 result = vma->vm_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382 current->mm->total_vm += len >> PAGE_SHIFT;
1383
David Howells8feae132009-01-08 12:04:47 +00001384share:
1385 add_vma_to_mm(current->mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386
Mike Frysingercfe79c02010-01-06 17:23:23 +00001387 /* we flush the region from the icache only when the first executable
1388 * mapping of it is made */
1389 if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1390 flush_icache_range(region->vm_start, region->vm_end);
1391 region->vm_icache_flushed = true;
1392 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393
Mike Frysingercfe79c02010-01-06 17:23:23 +00001394 up_write(&nommu_region_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395
David Howells8feae132009-01-08 12:04:47 +00001396 return result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397
David Howells8feae132009-01-08 12:04:47 +00001398error_just_free:
1399 up_write(&nommu_region_sem);
1400error:
David Howells89a86402009-10-30 13:13:26 +00001401 if (region->vm_file)
1402 fput(region->vm_file);
David Howells8feae132009-01-08 12:04:47 +00001403 kmem_cache_free(vm_region_jar, region);
David Howells89a86402009-10-30 13:13:26 +00001404 if (vma->vm_file)
1405 fput(vma->vm_file);
David Howells8feae132009-01-08 12:04:47 +00001406 kmem_cache_free(vm_area_cachep, vma);
David Howells8feae132009-01-08 12:04:47 +00001407 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408
David Howells8feae132009-01-08 12:04:47 +00001409sharing_violation:
1410 up_write(&nommu_region_sem);
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001411 pr_warn("Attempt to share mismatched mappings\n");
David Howells8feae132009-01-08 12:04:47 +00001412 ret = -EINVAL;
1413 goto error;
1414
1415error_getting_vma:
1416 kmem_cache_free(vm_region_jar, region);
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001417 pr_warn("Allocation of vma for %lu byte allocation from process %d failed\n",
1418 len, current->pid);
David Rientjes7bf02ea2011-05-24 17:11:16 -07001419 show_free_areas(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420 return -ENOMEM;
1421
David Howells8feae132009-01-08 12:04:47 +00001422error_getting_region:
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001423 pr_warn("Allocation of vm region for %lu byte allocation from process %d failed\n",
1424 len, current->pid);
David Rientjes7bf02ea2011-05-24 17:11:16 -07001425 show_free_areas(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 return -ENOMEM;
1427}
Linus Torvalds6be5ceb2012-04-20 17:13:58 -07001428
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001429SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1430 unsigned long, prot, unsigned long, flags,
1431 unsigned long, fd, unsigned long, pgoff)
1432{
1433 struct file *file = NULL;
1434 unsigned long retval = -EBADF;
1435
Al Viro120a7952010-10-30 02:54:44 -04001436 audit_mmap_fd(fd, flags);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001437 if (!(flags & MAP_ANONYMOUS)) {
1438 file = fget(fd);
1439 if (!file)
1440 goto out;
1441 }
1442
1443 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1444
Greg Ungererad1ed292012-06-04 14:29:59 +10001445 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001446
1447 if (file)
1448 fput(file);
1449out:
1450 return retval;
1451}
1452
Christoph Hellwiga4679372010-03-10 15:21:15 -08001453#ifdef __ARCH_WANT_SYS_OLD_MMAP
1454struct mmap_arg_struct {
1455 unsigned long addr;
1456 unsigned long len;
1457 unsigned long prot;
1458 unsigned long flags;
1459 unsigned long fd;
1460 unsigned long offset;
1461};
1462
1463SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1464{
1465 struct mmap_arg_struct a;
1466
1467 if (copy_from_user(&a, arg, sizeof(a)))
1468 return -EFAULT;
Alexander Kuleshov1824cb72015-11-05 18:46:35 -08001469 if (offset_in_page(a.offset))
Christoph Hellwiga4679372010-03-10 15:21:15 -08001470 return -EINVAL;
1471
1472 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1473 a.offset >> PAGE_SHIFT);
1474}
1475#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1476
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477/*
David Howells8feae132009-01-08 12:04:47 +00001478 * split a vma into two pieces at address 'addr', a new vma is allocated either
1479 * for the first part or the tail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 */
David Howells8feae132009-01-08 12:04:47 +00001481int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1482 unsigned long addr, int new_below)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483{
David Howells8feae132009-01-08 12:04:47 +00001484 struct vm_area_struct *new;
1485 struct vm_region *region;
1486 unsigned long npages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487
David Howells779c1022010-01-15 17:01:34 -08001488 /* we're only permitted to split anonymous regions (these should have
1489 * only a single usage on the region) */
1490 if (vma->vm_file)
David Howells8feae132009-01-08 12:04:47 +00001491 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492
David Howells8feae132009-01-08 12:04:47 +00001493 if (mm->map_count >= sysctl_max_map_count)
1494 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495
David Howells8feae132009-01-08 12:04:47 +00001496 region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1497 if (!region)
1498 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499
David Howells8feae132009-01-08 12:04:47 +00001500 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1501 if (!new) {
1502 kmem_cache_free(vm_region_jar, region);
1503 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 }
David Howells8feae132009-01-08 12:04:47 +00001505
1506 /* most fields are the same, copy all, and then fixup */
1507 *new = *vma;
1508 *region = *vma->vm_region;
1509 new->vm_region = region;
1510
1511 npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1512
1513 if (new_below) {
Paul Mundtdd8632a2009-01-08 12:04:47 +00001514 region->vm_top = region->vm_end = new->vm_end = addr;
David Howells8feae132009-01-08 12:04:47 +00001515 } else {
1516 region->vm_start = new->vm_start = addr;
1517 region->vm_pgoff = new->vm_pgoff += npages;
1518 }
1519
1520 if (new->vm_ops && new->vm_ops->open)
1521 new->vm_ops->open(new);
1522
1523 delete_vma_from_mm(vma);
1524 down_write(&nommu_region_sem);
1525 delete_nommu_region(vma->vm_region);
1526 if (new_below) {
1527 vma->vm_region->vm_start = vma->vm_start = addr;
1528 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1529 } else {
1530 vma->vm_region->vm_end = vma->vm_end = addr;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001531 vma->vm_region->vm_top = addr;
David Howells8feae132009-01-08 12:04:47 +00001532 }
1533 add_nommu_region(vma->vm_region);
1534 add_nommu_region(new->vm_region);
1535 up_write(&nommu_region_sem);
1536 add_vma_to_mm(mm, vma);
1537 add_vma_to_mm(mm, new);
1538 return 0;
1539}
1540
1541/*
1542 * shrink a VMA by removing the specified chunk from either the beginning or
1543 * the end
1544 */
1545static int shrink_vma(struct mm_struct *mm,
1546 struct vm_area_struct *vma,
1547 unsigned long from, unsigned long to)
1548{
1549 struct vm_region *region;
1550
David Howells8feae132009-01-08 12:04:47 +00001551 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1552 * and list */
1553 delete_vma_from_mm(vma);
1554 if (from > vma->vm_start)
1555 vma->vm_end = from;
1556 else
1557 vma->vm_start = to;
1558 add_vma_to_mm(mm, vma);
1559
1560 /* cut the backing region down to size */
1561 region = vma->vm_region;
David Howells1e2ae592010-01-15 17:01:33 -08001562 BUG_ON(region->vm_usage != 1);
David Howells8feae132009-01-08 12:04:47 +00001563
1564 down_write(&nommu_region_sem);
1565 delete_nommu_region(region);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001566 if (from > region->vm_start) {
1567 to = region->vm_top;
1568 region->vm_top = region->vm_end = from;
1569 } else {
David Howells8feae132009-01-08 12:04:47 +00001570 region->vm_start = to;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001571 }
David Howells8feae132009-01-08 12:04:47 +00001572 add_nommu_region(region);
1573 up_write(&nommu_region_sem);
1574
1575 free_page_series(from, to);
1576 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577}
1578
David Howells30340972006-09-27 01:50:20 -07001579/*
1580 * release a mapping
David Howells8feae132009-01-08 12:04:47 +00001581 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1582 * VMA, though it need not cover the whole VMA
David Howells30340972006-09-27 01:50:20 -07001583 */
David Howells8feae132009-01-08 12:04:47 +00001584int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585{
David Howells8feae132009-01-08 12:04:47 +00001586 struct vm_area_struct *vma;
Bob Liuf67d9b12011-05-24 17:12:56 -07001587 unsigned long end;
David Howells8feae132009-01-08 12:04:47 +00001588 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589
Bob Liuf67d9b12011-05-24 17:12:56 -07001590 len = PAGE_ALIGN(len);
David Howells8feae132009-01-08 12:04:47 +00001591 if (len == 0)
1592 return -EINVAL;
1593
Bob Liuf67d9b12011-05-24 17:12:56 -07001594 end = start + len;
1595
David Howells8feae132009-01-08 12:04:47 +00001596 /* find the first potentially overlapping VMA */
1597 vma = find_vma(mm, start);
1598 if (!vma) {
Choi Gi-yongac714902014-04-07 15:37:36 -07001599 static int limit;
David Howells33e5d7692009-04-02 16:56:32 -07001600 if (limit < 5) {
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001601 pr_warn("munmap of memory not mmapped by process %d (%s): 0x%lx-0x%lx\n",
1602 current->pid, current->comm,
1603 start, start + len - 1);
David Howells33e5d7692009-04-02 16:56:32 -07001604 limit++;
1605 }
David Howells8feae132009-01-08 12:04:47 +00001606 return -EINVAL;
David Howells30340972006-09-27 01:50:20 -07001607 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608
David Howells8feae132009-01-08 12:04:47 +00001609 /* we're allowed to split an anonymous VMA but not a file-backed one */
1610 if (vma->vm_file) {
1611 do {
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001612 if (start > vma->vm_start)
David Howells8feae132009-01-08 12:04:47 +00001613 return -EINVAL;
David Howells8feae132009-01-08 12:04:47 +00001614 if (end == vma->vm_end)
1615 goto erase_whole_vma;
Namhyung Kimd75a3102011-05-24 17:11:25 -07001616 vma = vma->vm_next;
1617 } while (vma);
David Howells8feae132009-01-08 12:04:47 +00001618 return -EINVAL;
1619 } else {
1620 /* the chunk must be a subset of the VMA found */
1621 if (start == vma->vm_start && end == vma->vm_end)
1622 goto erase_whole_vma;
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001623 if (start < vma->vm_start || end > vma->vm_end)
David Howells8feae132009-01-08 12:04:47 +00001624 return -EINVAL;
Alexander Kuleshov1824cb72015-11-05 18:46:35 -08001625 if (offset_in_page(start))
David Howells8feae132009-01-08 12:04:47 +00001626 return -EINVAL;
Alexander Kuleshov1824cb72015-11-05 18:46:35 -08001627 if (end != vma->vm_end && offset_in_page(end))
David Howells8feae132009-01-08 12:04:47 +00001628 return -EINVAL;
David Howells8feae132009-01-08 12:04:47 +00001629 if (start != vma->vm_start && end != vma->vm_end) {
1630 ret = split_vma(mm, vma, start, 1);
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001631 if (ret < 0)
David Howells8feae132009-01-08 12:04:47 +00001632 return ret;
David Howells8feae132009-01-08 12:04:47 +00001633 }
1634 return shrink_vma(mm, vma, start, end);
1635 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636
David Howells8feae132009-01-08 12:04:47 +00001637erase_whole_vma:
1638 delete_vma_from_mm(vma);
1639 delete_vma(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 return 0;
1641}
Paul Mundtb5073172007-07-21 04:37:25 -07001642EXPORT_SYMBOL(do_munmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643
Al Virobfce2812012-04-20 21:57:04 -04001644int vm_munmap(unsigned long addr, size_t len)
David Howells30340972006-09-27 01:50:20 -07001645{
Al Virobfce2812012-04-20 21:57:04 -04001646 struct mm_struct *mm = current->mm;
David Howells30340972006-09-27 01:50:20 -07001647 int ret;
David Howells30340972006-09-27 01:50:20 -07001648
1649 down_write(&mm->mmap_sem);
1650 ret = do_munmap(mm, addr, len);
1651 up_write(&mm->mmap_sem);
1652 return ret;
1653}
Linus Torvaldsa46ef992012-04-20 16:20:01 -07001654EXPORT_SYMBOL(vm_munmap);
1655
1656SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1657{
Al Virobfce2812012-04-20 21:57:04 -04001658 return vm_munmap(addr, len);
Linus Torvaldsa46ef992012-04-20 16:20:01 -07001659}
David Howells30340972006-09-27 01:50:20 -07001660
1661/*
David Howells8feae132009-01-08 12:04:47 +00001662 * release all the mappings made in a process's VM space
David Howells30340972006-09-27 01:50:20 -07001663 */
David Howells8feae132009-01-08 12:04:47 +00001664void exit_mmap(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665{
David Howells8feae132009-01-08 12:04:47 +00001666 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667
David Howells8feae132009-01-08 12:04:47 +00001668 if (!mm)
1669 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670
David Howells8feae132009-01-08 12:04:47 +00001671 mm->total_vm = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672
David Howells8feae132009-01-08 12:04:47 +00001673 while ((vma = mm->mmap)) {
1674 mm->mmap = vma->vm_next;
1675 delete_vma_from_mm(vma);
1676 delete_vma(mm, vma);
Steven J. Magnani04c34962010-11-24 12:56:54 -08001677 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678 }
1679}
1680
Linus Torvalds5d22fc22016-05-27 15:57:31 -07001681int vm_brk(unsigned long addr, unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682{
1683 return -ENOMEM;
1684}
1685
1686/*
David Howells6fa5f802006-09-27 01:50:21 -07001687 * expand (or shrink) an existing mapping, potentially moving it at the same
1688 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689 *
David Howells6fa5f802006-09-27 01:50:21 -07001690 * under NOMMU conditions, we only permit changing a mapping's size, and only
David Howells8feae132009-01-08 12:04:47 +00001691 * as long as it stays within the region allocated by do_mmap_private() and the
1692 * block is not shareable
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693 *
David Howells6fa5f802006-09-27 01:50:21 -07001694 * MREMAP_FIXED is not supported under NOMMU conditions
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695 */
Al Viro4b377ba2013-03-04 10:47:59 -05001696static unsigned long do_mremap(unsigned long addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 unsigned long old_len, unsigned long new_len,
1698 unsigned long flags, unsigned long new_addr)
1699{
David Howells6fa5f802006-09-27 01:50:21 -07001700 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701
1702 /* insanity checks first */
Bob Liuf67d9b12011-05-24 17:12:56 -07001703 old_len = PAGE_ALIGN(old_len);
1704 new_len = PAGE_ALIGN(new_len);
David Howells8feae132009-01-08 12:04:47 +00001705 if (old_len == 0 || new_len == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706 return (unsigned long) -EINVAL;
1707
Alexander Kuleshov1824cb72015-11-05 18:46:35 -08001708 if (offset_in_page(addr))
David Howells8feae132009-01-08 12:04:47 +00001709 return -EINVAL;
1710
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711 if (flags & MREMAP_FIXED && new_addr != addr)
1712 return (unsigned long) -EINVAL;
1713
David Howells8feae132009-01-08 12:04:47 +00001714 vma = find_vma_exact(current->mm, addr, old_len);
David Howells6fa5f802006-09-27 01:50:21 -07001715 if (!vma)
1716 return (unsigned long) -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717
David Howells6fa5f802006-09-27 01:50:21 -07001718 if (vma->vm_end != vma->vm_start + old_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719 return (unsigned long) -EFAULT;
1720
David Howells6fa5f802006-09-27 01:50:21 -07001721 if (vma->vm_flags & VM_MAYSHARE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722 return (unsigned long) -EPERM;
1723
David Howells8feae132009-01-08 12:04:47 +00001724 if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 return (unsigned long) -ENOMEM;
1726
1727 /* all checks complete - do it */
David Howells6fa5f802006-09-27 01:50:21 -07001728 vma->vm_end = vma->vm_start + new_len;
David Howells6fa5f802006-09-27 01:50:21 -07001729 return vma->vm_start;
1730}
1731
Heiko Carstens6a6160a2009-01-14 14:14:15 +01001732SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1733 unsigned long, new_len, unsigned long, flags,
1734 unsigned long, new_addr)
David Howells6fa5f802006-09-27 01:50:21 -07001735{
1736 unsigned long ret;
1737
1738 down_write(&current->mm->mmap_sem);
1739 ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1740 up_write(&current->mm->mmap_sem);
1741 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742}
1743
Michel Lespinasse240aade2013-02-22 16:35:56 -08001744struct page *follow_page_mask(struct vm_area_struct *vma,
1745 unsigned long address, unsigned int flags,
1746 unsigned int *page_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747{
Michel Lespinasse240aade2013-02-22 16:35:56 -08001748 *page_mask = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749 return NULL;
1750}
1751
Bob Liu8f3b1322011-07-08 15:39:46 -07001752int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1753 unsigned long pfn, unsigned long size, pgprot_t prot)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754{
Bob Liu8f3b1322011-07-08 15:39:46 -07001755 if (addr != (pfn << PAGE_SHIFT))
1756 return -EINVAL;
1757
Konstantin Khlebnikov314e51b2012-10-08 16:29:02 -07001758 vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
Greg Ungerer66aa2b42005-09-12 11:18:10 +10001759 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760}
Luke Yang22c4af42006-07-14 00:24:09 -07001761EXPORT_SYMBOL(remap_pfn_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762
Linus Torvalds3c0b9de2013-04-27 13:25:38 -07001763int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)
1764{
1765 unsigned long pfn = start >> PAGE_SHIFT;
1766 unsigned long vm_len = vma->vm_end - vma->vm_start;
1767
1768 pfn += vma->vm_pgoff;
1769 return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);
1770}
1771EXPORT_SYMBOL(vm_iomap_memory);
1772
Paul Mundtf905bc42008-02-04 22:29:59 -08001773int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1774 unsigned long pgoff)
1775{
1776 unsigned int size = vma->vm_end - vma->vm_start;
1777
1778 if (!(vma->vm_flags & VM_USERMAP))
1779 return -EINVAL;
1780
1781 vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1782 vma->vm_end = vma->vm_start + size;
1783
1784 return 0;
1785}
1786EXPORT_SYMBOL(remap_vmalloc_range);
1787
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1789 unsigned long len, unsigned long pgoff, unsigned long flags)
1790{
1791 return -ENOMEM;
1792}
1793
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794void unmap_mapping_range(struct address_space *mapping,
1795 loff_t const holebegin, loff_t const holelen,
1796 int even_cows)
1797{
1798}
Luke Yang22c4af42006-07-14 00:24:09 -07001799EXPORT_SYMBOL(unmap_mapping_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800
Nick Piggind0217ac2007-07-19 01:47:03 -07001801int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
David Howellsb0e15192006-01-06 00:11:42 -08001802{
1803 BUG();
Nick Piggind0217ac2007-07-19 01:47:03 -07001804 return 0;
David Howellsb0e15192006-01-06 00:11:42 -08001805}
Paul Mundtb5073172007-07-21 04:37:25 -07001806EXPORT_SYMBOL(filemap_fault);
David Howells0ec76a12006-09-27 01:50:15 -07001807
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07001808void filemap_map_pages(struct fault_env *fe,
1809 pgoff_t start_pgoff, pgoff_t end_pgoff)
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07001810{
1811 BUG();
1812}
1813EXPORT_SYMBOL(filemap_map_pages);
1814
Eric W. Biedermane71b4e02016-11-22 12:06:50 -06001815int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
Lorenzo Stoakes442486e2016-10-13 01:20:18 +01001816 unsigned long addr, void *buf, int len, unsigned int gup_flags)
David Howells0ec76a12006-09-27 01:50:15 -07001817{
David Howells0ec76a12006-09-27 01:50:15 -07001818 struct vm_area_struct *vma;
Lorenzo Stoakes442486e2016-10-13 01:20:18 +01001819 int write = gup_flags & FOLL_WRITE;
David Howells0ec76a12006-09-27 01:50:15 -07001820
1821 down_read(&mm->mmap_sem);
1822
1823 /* the access must start within one of the target process's mappings */
David Howells0159b142006-09-27 01:50:16 -07001824 vma = find_vma(mm, addr);
1825 if (vma) {
David Howells0ec76a12006-09-27 01:50:15 -07001826 /* don't overrun this mapping */
1827 if (addr + len >= vma->vm_end)
1828 len = vma->vm_end - addr;
1829
1830 /* only read or write mappings where it is permitted */
David Howellsd00c7b992006-09-27 01:50:19 -07001831 if (write && vma->vm_flags & VM_MAYWRITE)
Jie Zhang79597222010-01-06 17:23:28 +00001832 copy_to_user_page(vma, NULL, addr,
1833 (void *) addr, buf, len);
David Howellsd00c7b992006-09-27 01:50:19 -07001834 else if (!write && vma->vm_flags & VM_MAYREAD)
Jie Zhang79597222010-01-06 17:23:28 +00001835 copy_from_user_page(vma, NULL, addr,
1836 buf, (void *) addr, len);
David Howells0ec76a12006-09-27 01:50:15 -07001837 else
1838 len = 0;
1839 } else {
1840 len = 0;
1841 }
1842
1843 up_read(&mm->mmap_sem);
Mike Frysingerf55f1992011-03-29 14:05:12 +01001844
1845 return len;
1846}
1847
1848/**
1849 * @access_remote_vm - access another process' address space
1850 * @mm: the mm_struct of the target address space
1851 * @addr: start address to access
1852 * @buf: source or destination buffer
1853 * @len: number of bytes to transfer
Lorenzo Stoakes6347e8d2016-10-13 01:20:19 +01001854 * @gup_flags: flags modifying lookup behaviour
Mike Frysingerf55f1992011-03-29 14:05:12 +01001855 *
1856 * The caller must hold a reference on @mm.
1857 */
1858int access_remote_vm(struct mm_struct *mm, unsigned long addr,
Lorenzo Stoakes6347e8d2016-10-13 01:20:19 +01001859 void *buf, int len, unsigned int gup_flags)
Mike Frysingerf55f1992011-03-29 14:05:12 +01001860{
Lorenzo Stoakes6347e8d2016-10-13 01:20:19 +01001861 return __access_remote_vm(NULL, mm, addr, buf, len, gup_flags);
Mike Frysingerf55f1992011-03-29 14:05:12 +01001862}
1863
1864/*
1865 * Access another process' address space.
1866 * - source/target buffer must be kernel space
1867 */
Lorenzo Stoakesf307ab62016-10-13 01:20:20 +01001868int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len,
1869 unsigned int gup_flags)
Mike Frysingerf55f1992011-03-29 14:05:12 +01001870{
1871 struct mm_struct *mm;
1872
1873 if (addr + len < addr)
1874 return 0;
1875
1876 mm = get_task_mm(tsk);
1877 if (!mm)
1878 return 0;
1879
Lorenzo Stoakesf307ab62016-10-13 01:20:20 +01001880 len = __access_remote_vm(tsk, mm, addr, buf, len, gup_flags);
Mike Frysingerf55f1992011-03-29 14:05:12 +01001881
David Howells0ec76a12006-09-27 01:50:15 -07001882 mmput(mm);
1883 return len;
1884}
David Howells7e660872010-01-15 17:01:39 -08001885
1886/**
1887 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
1888 * @inode: The inode to check
1889 * @size: The current filesize of the inode
1890 * @newsize: The proposed filesize of the inode
1891 *
1892 * Check the shared mappings on an inode on behalf of a shrinking truncate to
1893 * make sure that that any outstanding VMAs aren't broken and then shrink the
1894 * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
1895 * automatically grant mappings that are too large.
1896 */
1897int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
1898 size_t newsize)
1899{
1900 struct vm_area_struct *vma;
David Howells7e660872010-01-15 17:01:39 -08001901 struct vm_region *region;
1902 pgoff_t low, high;
1903 size_t r_size, r_top;
1904
1905 low = newsize >> PAGE_SHIFT;
1906 high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1907
1908 down_write(&nommu_region_sem);
Davidlohr Bueso1acf2e02014-12-12 16:54:39 -08001909 i_mmap_lock_read(inode->i_mapping);
David Howells7e660872010-01-15 17:01:39 -08001910
1911 /* search for VMAs that fall within the dead zone */
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -07001912 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) {
David Howells7e660872010-01-15 17:01:39 -08001913 /* found one - only interested if it's shared out of the page
1914 * cache */
1915 if (vma->vm_flags & VM_SHARED) {
Davidlohr Bueso1acf2e02014-12-12 16:54:39 -08001916 i_mmap_unlock_read(inode->i_mapping);
David Howells7e660872010-01-15 17:01:39 -08001917 up_write(&nommu_region_sem);
1918 return -ETXTBSY; /* not quite true, but near enough */
1919 }
1920 }
1921
1922 /* reduce any regions that overlap the dead zone - if in existence,
1923 * these will be pointed to by VMAs that don't overlap the dead zone
1924 *
1925 * we don't check for any regions that start beyond the EOF as there
1926 * shouldn't be any
1927 */
Davidlohr Bueso1acf2e02014-12-12 16:54:39 -08001928 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, 0, ULONG_MAX) {
David Howells7e660872010-01-15 17:01:39 -08001929 if (!(vma->vm_flags & VM_SHARED))
1930 continue;
1931
1932 region = vma->vm_region;
1933 r_size = region->vm_top - region->vm_start;
1934 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
1935
1936 if (r_top > newsize) {
1937 region->vm_top -= r_top - newsize;
1938 if (region->vm_end > region->vm_top)
1939 region->vm_end = region->vm_top;
1940 }
1941 }
1942
Davidlohr Bueso1acf2e02014-12-12 16:54:39 -08001943 i_mmap_unlock_read(inode->i_mapping);
David Howells7e660872010-01-15 17:01:39 -08001944 up_write(&nommu_region_sem);
1945 return 0;
1946}
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07001947
1948/*
1949 * Initialise sysctl_user_reserve_kbytes.
1950 *
1951 * This is intended to prevent a user from starting a single memory hogging
1952 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
1953 * mode.
1954 *
1955 * The default value is min(3% of free memory, 128MB)
1956 * 128MB is enough to recover with sshd/login, bash, and top/kill.
1957 */
1958static int __meminit init_user_reserve(void)
1959{
1960 unsigned long free_kbytes;
1961
1962 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
1963
1964 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
1965 return 0;
1966}
Paul Gortmakera4bc6fc2015-05-01 20:08:20 -04001967subsys_initcall(init_user_reserve);
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07001968
1969/*
1970 * Initialise sysctl_admin_reserve_kbytes.
1971 *
1972 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
1973 * to log in and kill a memory hogging process.
1974 *
1975 * Systems with more than 256MB will reserve 8MB, enough to recover
1976 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
1977 * only reserve 3% of free pages by default.
1978 */
1979static int __meminit init_admin_reserve(void)
1980{
1981 unsigned long free_kbytes;
1982
1983 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
1984
1985 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
1986 return 0;
1987}
Paul Gortmakera4bc6fc2015-05-01 20:08:20 -04001988subsys_initcall(init_admin_reserve);