blob: de8b6b6580c1eddaeca1ab2ee05350e5a6b31620 [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
Dave Hansencde70142016-02-12 13:01:55 -080018#define __DISABLE_GUP_DEPRECATED
19
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040020#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include <linux/mm.h>
Davidlohr Bueso615d6e82014-04-07 15:37:25 -070022#include <linux/vmacache.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <linux/mman.h>
24#include <linux/swap.h>
25#include <linux/file.h>
26#include <linux/highmem.h>
27#include <linux/pagemap.h>
28#include <linux/slab.h>
29#include <linux/vmalloc.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include <linux/blkdev.h>
31#include <linux/backing-dev.h>
Gideon Israel Dsouza3b321232014-04-07 15:37:26 -070032#include <linux/compiler.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include <linux/mount.h>
34#include <linux/personality.h>
35#include <linux/security.h>
36#include <linux/syscalls.h>
Al Viro120a7952010-10-30 02:54:44 -040037#include <linux/audit.h>
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -070038#include <linux/printk.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070039
40#include <asm/uaccess.h>
41#include <asm/tlb.h>
42#include <asm/tlbflush.h>
Bernd Schmidteb8cdec2009-09-21 17:03:57 -070043#include <asm/mmu_context.h>
David Howells8feae132009-01-08 12:04:47 +000044#include "internal.h"
45
Linus Torvalds1da177e2005-04-16 15:20:36 -070046void *high_memory;
Arnd Bergmann944b6872015-02-05 12:25:12 -080047EXPORT_SYMBOL(high_memory);
Linus Torvalds1da177e2005-04-16 15:20:36 -070048struct page *mem_map;
49unsigned long max_mapnr;
gchen gchen5b8bf302015-03-12 16:26:05 -070050EXPORT_SYMBOL(max_mapnr);
Hugh Dickins4266c972009-09-23 17:05:53 +010051unsigned long highest_memmap_pfn;
David Howellsfc4d5c22009-05-06 16:03:05 -070052int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -070053int heap_stack_gap = 0;
54
David Howells33e5d7692009-04-02 16:56:32 -070055atomic_long_t mmap_pages_allocated;
David Howells8feae132009-01-08 12:04:47 +000056
Linus Torvalds1da177e2005-04-16 15:20:36 -070057EXPORT_SYMBOL(mem_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -070058
David Howells8feae132009-01-08 12:04:47 +000059/* list of mapped, potentially shareable regions */
60static struct kmem_cache *vm_region_jar;
61struct rb_root nommu_region_tree = RB_ROOT;
62DECLARE_RWSEM(nommu_region_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -070063
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +040064const struct vm_operations_struct generic_file_vm_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -070065};
66
67/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070068 * Return the total memory allocated for this pointer, not
69 * just what the caller asked for.
70 *
71 * Doesn't have to be accurate, i.e. may have races.
72 */
73unsigned int kobjsize(const void *objp)
74{
75 struct page *page;
76
Michael Hennerich4016a132008-04-28 02:13:38 -070077 /*
78 * If the object we have should not have ksize performed on it,
79 * return size of 0
80 */
Paul Mundt5a1603b2008-06-12 16:29:55 +090081 if (!objp || !virt_addr_valid(objp))
Paul Mundt6cfd53fc2008-06-05 22:46:08 -070082 return 0;
83
84 page = virt_to_head_page(objp);
Paul Mundt6cfd53fc2008-06-05 22:46:08 -070085
86 /*
87 * If the allocator sets PageSlab, we know the pointer came from
88 * kmalloc().
89 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070090 if (PageSlab(page))
91 return ksize(objp);
92
Paul Mundt6cfd53fc2008-06-05 22:46:08 -070093 /*
Paul Mundtab2e83e2009-01-08 12:04:48 +000094 * If it's not a compound page, see if we have a matching VMA
95 * region. This test is intentionally done in reverse order,
96 * so if there's no VMA, we still fall through and hand back
97 * PAGE_SIZE for 0-order pages.
98 */
99 if (!PageCompound(page)) {
100 struct vm_area_struct *vma;
101
102 vma = find_vma(current->mm, (unsigned long)objp);
103 if (vma)
104 return vma->vm_end - vma->vm_start;
105 }
106
107 /*
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700108 * The ksize() function is only guaranteed to work for pointers
Paul Mundt5a1603b2008-06-12 16:29:55 +0900109 * returned by kmalloc(). So handle arbitrary pointers here.
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700110 */
Paul Mundt5a1603b2008-06-12 16:29:55 +0900111 return PAGE_SIZE << compound_order(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112}
113
Michel Lespinasse28a35712013-02-22 16:35:55 -0800114long __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
115 unsigned long start, unsigned long nr_pages,
116 unsigned int foll_flags, struct page **pages,
117 struct vm_area_struct **vmas, int *nonblocking)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118{
Sonic Zhang910e46d2006-09-27 01:50:17 -0700119 struct vm_area_struct *vma;
David Howells7b4d5b82006-09-27 01:50:18 -0700120 unsigned long vm_flags;
121 int i;
122
123 /* calculate required read or write permissions.
Hugh Dickins58fa8792009-09-21 17:03:31 -0700124 * If FOLL_FORCE is set, we only require the "MAY" flags.
David Howells7b4d5b82006-09-27 01:50:18 -0700125 */
Hugh Dickins58fa8792009-09-21 17:03:31 -0700126 vm_flags = (foll_flags & FOLL_WRITE) ?
127 (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
128 vm_flags &= (foll_flags & FOLL_FORCE) ?
129 (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130
Peter Zijlstra9d737772009-06-25 11:58:55 +0200131 for (i = 0; i < nr_pages; i++) {
David Howells7561e8c2010-03-25 16:48:38 +0000132 vma = find_vma(mm, start);
David Howells7b4d5b82006-09-27 01:50:18 -0700133 if (!vma)
134 goto finish_or_fault;
135
136 /* protect what we can, including chardevs */
Hugh Dickins1c3aff12009-09-21 17:03:24 -0700137 if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
138 !(vm_flags & vma->vm_flags))
David Howells7b4d5b82006-09-27 01:50:18 -0700139 goto finish_or_fault;
Sonic Zhang910e46d2006-09-27 01:50:17 -0700140
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141 if (pages) {
142 pages[i] = virt_to_page(start);
143 if (pages[i])
144 page_cache_get(pages[i]);
145 }
146 if (vmas)
Sonic Zhang910e46d2006-09-27 01:50:17 -0700147 vmas[i] = vma;
David Howellse1ee65d2010-03-25 16:48:44 +0000148 start = (start + PAGE_SIZE) & PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 }
David Howells7b4d5b82006-09-27 01:50:18 -0700150
151 return i;
152
153finish_or_fault:
154 return i ? : -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155}
Nick Pigginb291f002008-10-18 20:26:44 -0700156
Nick Pigginb291f002008-10-18 20:26:44 -0700157/*
158 * get a list of pages in an address range belonging to the specified process
159 * and indicate the VMA that covers each page
160 * - this is potentially dodgy as we may end incrementing the page count of a
161 * slab page or a secondary page from a compound page
162 * - don't permit access to VMAs that don't support it, such as I/O mappings
163 */
Dave Hansencde70142016-02-12 13:01:55 -0800164long get_user_pages6(unsigned long start, unsigned long nr_pages,
Michel Lespinasse28a35712013-02-22 16:35:55 -0800165 int write, int force, struct page **pages,
166 struct vm_area_struct **vmas)
Nick Pigginb291f002008-10-18 20:26:44 -0700167{
168 int flags = 0;
169
170 if (write)
Hugh Dickins58fa8792009-09-21 17:03:31 -0700171 flags |= FOLL_WRITE;
Nick Pigginb291f002008-10-18 20:26:44 -0700172 if (force)
Hugh Dickins58fa8792009-09-21 17:03:31 -0700173 flags |= FOLL_FORCE;
Nick Pigginb291f002008-10-18 20:26:44 -0700174
Dave Hansencde70142016-02-12 13:01:55 -0800175 return __get_user_pages(current, current->mm, start, nr_pages, flags,
176 pages, vmas, NULL);
Nick Pigginb291f002008-10-18 20:26:44 -0700177}
Dave Hansencde70142016-02-12 13:01:55 -0800178EXPORT_SYMBOL(get_user_pages6);
Greg Ungerer66aa2b42005-09-12 11:18:10 +1000179
Dave Hansencde70142016-02-12 13:01:55 -0800180long get_user_pages_locked6(unsigned long start, unsigned long nr_pages,
181 int write, int force, struct page **pages,
182 int *locked)
Andrea Arcangelif0818f42015-02-11 15:27:17 -0800183{
Dave Hansencde70142016-02-12 13:01:55 -0800184 return get_user_pages6(start, nr_pages, write, force, pages, NULL);
Andrea Arcangelif0818f42015-02-11 15:27:17 -0800185}
Dave Hansencde70142016-02-12 13:01:55 -0800186EXPORT_SYMBOL(get_user_pages_locked6);
Andrea Arcangelif0818f42015-02-11 15:27:17 -0800187
Andrea Arcangeli0fd71a52015-02-11 15:27:20 -0800188long __get_user_pages_unlocked(struct task_struct *tsk, struct mm_struct *mm,
189 unsigned long start, unsigned long nr_pages,
190 int write, int force, struct page **pages,
191 unsigned int gup_flags)
Andrea Arcangelif0818f42015-02-11 15:27:17 -0800192{
193 long ret;
194 down_read(&mm->mmap_sem);
Dave Hansencde70142016-02-12 13:01:55 -0800195 ret = __get_user_pages(tsk, mm, start, nr_pages, gup_flags, pages,
196 NULL, NULL);
Andrea Arcangelif0818f42015-02-11 15:27:17 -0800197 up_read(&mm->mmap_sem);
198 return ret;
199}
Andrea Arcangeli0fd71a52015-02-11 15:27:20 -0800200EXPORT_SYMBOL(__get_user_pages_unlocked);
201
Dave Hansencde70142016-02-12 13:01:55 -0800202long get_user_pages_unlocked5(unsigned long start, unsigned long nr_pages,
Andrea Arcangeli0fd71a52015-02-11 15:27:20 -0800203 int write, int force, struct page **pages)
204{
Dave Hansencde70142016-02-12 13:01:55 -0800205 return __get_user_pages_unlocked(current, current->mm, start, nr_pages,
206 write, force, pages, 0);
Andrea Arcangeli0fd71a52015-02-11 15:27:20 -0800207}
Dave Hansencde70142016-02-12 13:01:55 -0800208EXPORT_SYMBOL(get_user_pages_unlocked5);
Andrea Arcangelif0818f42015-02-11 15:27:17 -0800209
Paul Mundtdfc2f912009-06-26 04:31:57 +0900210/**
211 * follow_pfn - look up PFN at a user virtual address
212 * @vma: memory mapping
213 * @address: user virtual address
214 * @pfn: location to store found PFN
215 *
216 * Only IO mappings and raw PFN mappings are allowed.
217 *
218 * Returns zero and the pfn at @pfn on success, -ve otherwise.
219 */
220int follow_pfn(struct vm_area_struct *vma, unsigned long address,
221 unsigned long *pfn)
222{
223 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
224 return -EINVAL;
225
226 *pfn = address >> PAGE_SHIFT;
227 return 0;
228}
229EXPORT_SYMBOL(follow_pfn);
230
Joonsoo Kimf1c40692013-04-29 15:07:37 -0700231LIST_HEAD(vmap_area_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800233void vfree(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234{
235 kfree(addr);
236}
Paul Mundtb5073172007-07-21 04:37:25 -0700237EXPORT_SYMBOL(vfree);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238
Al Virodd0fc662005-10-07 07:46:04 +0100239void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240{
241 /*
Robert P. J. Day85186092007-10-19 23:11:38 +0200242 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
243 * returns only a logical address.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244 */
Nick Piggin84097512006-03-22 00:08:34 -0800245 return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246}
Paul Mundtb5073172007-07-21 04:37:25 -0700247EXPORT_SYMBOL(__vmalloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248
Paul Mundtf905bc42008-02-04 22:29:59 -0800249void *vmalloc_user(unsigned long size)
250{
251 void *ret;
252
253 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
254 PAGE_KERNEL);
255 if (ret) {
256 struct vm_area_struct *vma;
257
258 down_write(&current->mm->mmap_sem);
259 vma = find_vma(current->mm, (unsigned long)ret);
260 if (vma)
261 vma->vm_flags |= VM_USERMAP;
262 up_write(&current->mm->mmap_sem);
263 }
264
265 return ret;
266}
267EXPORT_SYMBOL(vmalloc_user);
268
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800269struct page *vmalloc_to_page(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270{
271 return virt_to_page(addr);
272}
Paul Mundtb5073172007-07-21 04:37:25 -0700273EXPORT_SYMBOL(vmalloc_to_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800275unsigned long vmalloc_to_pfn(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276{
277 return page_to_pfn(virt_to_page(addr));
278}
Paul Mundtb5073172007-07-21 04:37:25 -0700279EXPORT_SYMBOL(vmalloc_to_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280
281long vread(char *buf, char *addr, unsigned long count)
282{
Chen Gang9bde9162013-07-03 15:02:36 -0700283 /* Don't allow overflow */
284 if ((unsigned long) buf + count < count)
285 count = -(unsigned long) buf;
286
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 memcpy(buf, addr, count);
288 return count;
289}
290
291long vwrite(char *buf, char *addr, unsigned long count)
292{
293 /* Don't allow overflow */
294 if ((unsigned long) addr + count < count)
295 count = -(unsigned long) addr;
296
297 memcpy(addr, buf, count);
Choi Gi-yongac714902014-04-07 15:37:36 -0700298 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299}
300
301/*
Masahiro Yamadae1c05062015-07-07 10:14:59 +0900302 * vmalloc - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303 *
304 * @size: allocation size
305 *
306 * Allocate enough pages to cover @size from the page level
Masahiro Yamadae1c05062015-07-07 10:14:59 +0900307 * allocator and map them into contiguous kernel virtual space.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +0200309 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 * use __vmalloc() instead.
311 */
312void *vmalloc(unsigned long size)
313{
314 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
315}
Andrew Mortonf6138882006-02-28 16:59:18 -0800316EXPORT_SYMBOL(vmalloc);
317
Dave Younge1ca7782010-10-26 14:22:06 -0700318/*
Masahiro Yamadae1c05062015-07-07 10:14:59 +0900319 * vzalloc - allocate virtually contiguous memory with zero fill
Dave Younge1ca7782010-10-26 14:22:06 -0700320 *
321 * @size: allocation size
322 *
323 * Allocate enough pages to cover @size from the page level
Masahiro Yamadae1c05062015-07-07 10:14:59 +0900324 * allocator and map them into contiguous kernel virtual space.
Dave Younge1ca7782010-10-26 14:22:06 -0700325 * The memory allocated is set to zero.
326 *
327 * For tight control over page level allocator and protection flags
328 * use __vmalloc() instead.
329 */
330void *vzalloc(unsigned long size)
331{
332 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
333 PAGE_KERNEL);
334}
335EXPORT_SYMBOL(vzalloc);
336
337/**
338 * vmalloc_node - allocate memory on a specific node
339 * @size: allocation size
340 * @node: numa node
341 *
342 * Allocate enough pages to cover @size from the page level
343 * allocator and map them into contiguous kernel virtual space.
344 *
345 * For tight control over page level allocator and protection flags
346 * use __vmalloc() instead.
347 */
Andrew Mortonf6138882006-02-28 16:59:18 -0800348void *vmalloc_node(unsigned long size, int node)
349{
350 return vmalloc(size);
351}
Paul Mundt9a14f652010-12-24 11:50:34 +0900352EXPORT_SYMBOL(vmalloc_node);
Dave Younge1ca7782010-10-26 14:22:06 -0700353
354/**
355 * vzalloc_node - allocate memory on a specific node with zero fill
356 * @size: allocation size
357 * @node: numa node
358 *
359 * Allocate enough pages to cover @size from the page level
360 * allocator and map them into contiguous kernel virtual space.
361 * The memory allocated is set to zero.
362 *
363 * For tight control over page level allocator and protection flags
364 * use __vmalloc() instead.
365 */
366void *vzalloc_node(unsigned long size, int node)
367{
368 return vzalloc(size);
369}
370EXPORT_SYMBOL(vzalloc_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371
Paul Mundt1af446e2008-08-04 16:01:47 +0900372#ifndef PAGE_KERNEL_EXEC
373# define PAGE_KERNEL_EXEC PAGE_KERNEL
374#endif
375
376/**
377 * vmalloc_exec - allocate virtually contiguous, executable memory
378 * @size: allocation size
379 *
380 * Kernel-internal function to allocate enough pages to cover @size
381 * the page level allocator and map them into contiguous and
382 * executable kernel virtual space.
383 *
384 * For tight control over page level allocator and protection flags
385 * use __vmalloc() instead.
386 */
387
388void *vmalloc_exec(unsigned long size)
389{
390 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
391}
392
Paul Mundtb5073172007-07-21 04:37:25 -0700393/**
394 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 * @size: allocation size
396 *
397 * Allocate enough 32bit PA addressable pages to cover @size from the
Masahiro Yamadae1c05062015-07-07 10:14:59 +0900398 * page level allocator and map them into contiguous kernel virtual space.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 */
400void *vmalloc_32(unsigned long size)
401{
402 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
403}
Paul Mundtb5073172007-07-21 04:37:25 -0700404EXPORT_SYMBOL(vmalloc_32);
405
406/**
407 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
408 * @size: allocation size
409 *
410 * The resulting memory area is 32bit addressable and zeroed so it can be
411 * mapped to userspace without leaking data.
Paul Mundtf905bc42008-02-04 22:29:59 -0800412 *
413 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
414 * remap_vmalloc_range() are permissible.
Paul Mundtb5073172007-07-21 04:37:25 -0700415 */
416void *vmalloc_32_user(unsigned long size)
417{
Paul Mundtf905bc42008-02-04 22:29:59 -0800418 /*
419 * We'll have to sort out the ZONE_DMA bits for 64-bit,
420 * but for now this can simply use vmalloc_user() directly.
421 */
422 return vmalloc_user(size);
Paul Mundtb5073172007-07-21 04:37:25 -0700423}
424EXPORT_SYMBOL(vmalloc_32_user);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425
426void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
427{
428 BUG();
429 return NULL;
430}
Paul Mundtb5073172007-07-21 04:37:25 -0700431EXPORT_SYMBOL(vmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800433void vunmap(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434{
435 BUG();
436}
Paul Mundtb5073172007-07-21 04:37:25 -0700437EXPORT_SYMBOL(vunmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438
Paul Mundteb6434d2009-01-21 17:45:47 +0900439void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
440{
441 BUG();
442 return NULL;
443}
444EXPORT_SYMBOL(vm_map_ram);
445
446void vm_unmap_ram(const void *mem, unsigned int count)
447{
448 BUG();
449}
450EXPORT_SYMBOL(vm_unmap_ram);
451
452void vm_unmap_aliases(void)
453{
454}
455EXPORT_SYMBOL_GPL(vm_unmap_aliases);
456
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457/*
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -0700458 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
459 * have one.
460 */
Gideon Israel Dsouza3b321232014-04-07 15:37:26 -0700461void __weak vmalloc_sync_all(void)
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -0700462{
463}
464
Paul Mundt29c185e2010-12-24 12:08:30 +0900465/**
466 * alloc_vm_area - allocate a range of kernel address space
467 * @size: size of the area
468 *
469 * Returns: NULL on failure, vm_struct on success
470 *
471 * This function reserves a range of kernel address space, and
472 * allocates pagetables to map that range. No actual mappings
473 * are created. If the kernel address space is not shared
474 * between processes, it syncs the pagetable across all
475 * processes.
476 */
David Vrabelcd129092011-09-29 16:53:32 +0100477struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
Paul Mundt29c185e2010-12-24 12:08:30 +0900478{
479 BUG();
480 return NULL;
481}
482EXPORT_SYMBOL_GPL(alloc_vm_area);
483
484void free_vm_area(struct vm_struct *area)
485{
486 BUG();
487}
488EXPORT_SYMBOL_GPL(free_vm_area);
489
Paul Mundtb5073172007-07-21 04:37:25 -0700490int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
491 struct page *page)
492{
493 return -EINVAL;
494}
495EXPORT_SYMBOL(vm_insert_page);
496
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -0700497/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 * sys_brk() for the most part doesn't need the global kernel
499 * lock, except when an application is doing something nasty
500 * like trying to un-brk an area that has already been mapped
501 * to a regular file. in this case, the unmapping will need
502 * to invoke file system routines that need the global lock.
503 */
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100504SYSCALL_DEFINE1(brk, unsigned long, brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505{
506 struct mm_struct *mm = current->mm;
507
508 if (brk < mm->start_brk || brk > mm->context.end_brk)
509 return mm->brk;
510
511 if (mm->brk == brk)
512 return mm->brk;
513
514 /*
515 * Always allow shrinking brk
516 */
517 if (brk <= mm->brk) {
518 mm->brk = brk;
519 return brk;
520 }
521
522 /*
523 * Ok, looks good - let it rip.
524 */
Mike Frysingercfe79c02010-01-06 17:23:23 +0000525 flush_icache_range(mm->brk, brk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 return mm->brk = brk;
527}
528
David Howells8feae132009-01-08 12:04:47 +0000529/*
530 * initialise the VMA and region record slabs
531 */
532void __init mmap_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533{
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -0700534 int ret;
535
Tejun Heo908c7f12014-09-08 09:51:29 +0900536 ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -0700537 VM_BUG_ON(ret);
Vladimir Davydov5d097052016-01-14 15:18:21 -0800538 vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC|SLAB_ACCOUNT);
David Howells8feae132009-01-08 12:04:47 +0000539}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540
David Howells8feae132009-01-08 12:04:47 +0000541/*
542 * validate the region tree
543 * - the caller must hold the region lock
544 */
545#ifdef CONFIG_DEBUG_NOMMU_REGIONS
546static noinline void validate_nommu_regions(void)
547{
548 struct vm_region *region, *last;
549 struct rb_node *p, *lastp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550
David Howells8feae132009-01-08 12:04:47 +0000551 lastp = rb_first(&nommu_region_tree);
552 if (!lastp)
553 return;
554
555 last = rb_entry(lastp, struct vm_region, vm_rb);
Geliang Tangc9427bc2015-11-05 18:48:38 -0800556 BUG_ON(last->vm_end <= last->vm_start);
557 BUG_ON(last->vm_top < last->vm_end);
David Howells8feae132009-01-08 12:04:47 +0000558
559 while ((p = rb_next(lastp))) {
560 region = rb_entry(p, struct vm_region, vm_rb);
561 last = rb_entry(lastp, struct vm_region, vm_rb);
562
Geliang Tangc9427bc2015-11-05 18:48:38 -0800563 BUG_ON(region->vm_end <= region->vm_start);
564 BUG_ON(region->vm_top < region->vm_end);
565 BUG_ON(region->vm_start < last->vm_top);
David Howells8feae132009-01-08 12:04:47 +0000566
567 lastp = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 }
569}
David Howells8feae132009-01-08 12:04:47 +0000570#else
David Howells33e5d7692009-04-02 16:56:32 -0700571static void validate_nommu_regions(void)
572{
573}
David Howells8feae132009-01-08 12:04:47 +0000574#endif
575
576/*
577 * add a region into the global tree
578 */
579static void add_nommu_region(struct vm_region *region)
580{
581 struct vm_region *pregion;
582 struct rb_node **p, *parent;
583
584 validate_nommu_regions();
585
David Howells8feae132009-01-08 12:04:47 +0000586 parent = NULL;
587 p = &nommu_region_tree.rb_node;
588 while (*p) {
589 parent = *p;
590 pregion = rb_entry(parent, struct vm_region, vm_rb);
591 if (region->vm_start < pregion->vm_start)
592 p = &(*p)->rb_left;
593 else if (region->vm_start > pregion->vm_start)
594 p = &(*p)->rb_right;
595 else if (pregion == region)
596 return;
597 else
598 BUG();
599 }
600
601 rb_link_node(&region->vm_rb, parent, p);
602 rb_insert_color(&region->vm_rb, &nommu_region_tree);
603
604 validate_nommu_regions();
605}
606
607/*
608 * delete a region from the global tree
609 */
610static void delete_nommu_region(struct vm_region *region)
611{
612 BUG_ON(!nommu_region_tree.rb_node);
613
614 validate_nommu_regions();
615 rb_erase(&region->vm_rb, &nommu_region_tree);
616 validate_nommu_regions();
617}
618
619/*
620 * free a contiguous series of pages
621 */
622static void free_page_series(unsigned long from, unsigned long to)
623{
624 for (; from < to; from += PAGE_SIZE) {
625 struct page *page = virt_to_page(from);
626
David Howells33e5d7692009-04-02 16:56:32 -0700627 atomic_long_dec(&mmap_pages_allocated);
David Howells8feae132009-01-08 12:04:47 +0000628 put_page(page);
629 }
630}
631
632/*
633 * release a reference to a region
David Howells33e5d7692009-04-02 16:56:32 -0700634 * - the caller must hold the region semaphore for writing, which this releases
Paul Mundtdd8632a2009-01-08 12:04:47 +0000635 * - the region may not have been added to the tree yet, in which case vm_top
David Howells8feae132009-01-08 12:04:47 +0000636 * will equal vm_start
637 */
638static void __put_nommu_region(struct vm_region *region)
639 __releases(nommu_region_sem)
640{
David Howells8feae132009-01-08 12:04:47 +0000641 BUG_ON(!nommu_region_tree.rb_node);
642
David Howells1e2ae592010-01-15 17:01:33 -0800643 if (--region->vm_usage == 0) {
Paul Mundtdd8632a2009-01-08 12:04:47 +0000644 if (region->vm_top > region->vm_start)
David Howells8feae132009-01-08 12:04:47 +0000645 delete_nommu_region(region);
646 up_write(&nommu_region_sem);
647
648 if (region->vm_file)
649 fput(region->vm_file);
650
651 /* IO memory and memory shared directly out of the pagecache
652 * from ramfs/tmpfs mustn't be released here */
Leon Romanovsky22cc8772015-06-24 16:57:47 -0700653 if (region->vm_flags & VM_MAPPED_COPY)
Paul Mundtdd8632a2009-01-08 12:04:47 +0000654 free_page_series(region->vm_start, region->vm_top);
David Howells8feae132009-01-08 12:04:47 +0000655 kmem_cache_free(vm_region_jar, region);
656 } else {
657 up_write(&nommu_region_sem);
658 }
659}
660
661/*
662 * release a reference to a region
663 */
664static void put_nommu_region(struct vm_region *region)
665{
666 down_write(&nommu_region_sem);
667 __put_nommu_region(region);
668}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669
David Howells30340972006-09-27 01:50:20 -0700670/*
Bernd Schmidteb8cdec2009-09-21 17:03:57 -0700671 * update protection on a vma
672 */
673static void protect_vma(struct vm_area_struct *vma, unsigned long flags)
674{
675#ifdef CONFIG_MPU
676 struct mm_struct *mm = vma->vm_mm;
677 long start = vma->vm_start & PAGE_MASK;
678 while (start < vma->vm_end) {
679 protect_page(mm, start, flags);
680 start += PAGE_SIZE;
681 }
682 update_protections(mm);
683#endif
684}
685
686/*
David Howells30340972006-09-27 01:50:20 -0700687 * add a VMA into a process's mm_struct in the appropriate place in the list
David Howells8feae132009-01-08 12:04:47 +0000688 * and tree and add to the address space's page tree also if not an anonymous
689 * page
David Howells30340972006-09-27 01:50:20 -0700690 * - should be called with mm->mmap_sem held writelocked
691 */
David Howells8feae132009-01-08 12:04:47 +0000692static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
David Howells30340972006-09-27 01:50:20 -0700693{
Namhyung Kim6038def2011-05-24 17:11:22 -0700694 struct vm_area_struct *pvma, *prev;
David Howells8feae132009-01-08 12:04:47 +0000695 struct address_space *mapping;
Namhyung Kim6038def2011-05-24 17:11:22 -0700696 struct rb_node **p, *parent, *rb_prev;
David Howells30340972006-09-27 01:50:20 -0700697
David Howells8feae132009-01-08 12:04:47 +0000698 BUG_ON(!vma->vm_region);
699
700 mm->map_count++;
701 vma->vm_mm = mm;
702
Bernd Schmidteb8cdec2009-09-21 17:03:57 -0700703 protect_vma(vma, vma->vm_flags);
704
David Howells8feae132009-01-08 12:04:47 +0000705 /* add the VMA to the mapping */
706 if (vma->vm_file) {
707 mapping = vma->vm_file->f_mapping;
708
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800709 i_mmap_lock_write(mapping);
David Howells8feae132009-01-08 12:04:47 +0000710 flush_dcache_mmap_lock(mapping);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700711 vma_interval_tree_insert(vma, &mapping->i_mmap);
David Howells8feae132009-01-08 12:04:47 +0000712 flush_dcache_mmap_unlock(mapping);
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800713 i_mmap_unlock_write(mapping);
David Howells8feae132009-01-08 12:04:47 +0000714 }
715
716 /* add the VMA to the tree */
Namhyung Kim6038def2011-05-24 17:11:22 -0700717 parent = rb_prev = NULL;
David Howells8feae132009-01-08 12:04:47 +0000718 p = &mm->mm_rb.rb_node;
719 while (*p) {
720 parent = *p;
721 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
722
723 /* sort by: start addr, end addr, VMA struct addr in that order
724 * (the latter is necessary as we may get identical VMAs) */
725 if (vma->vm_start < pvma->vm_start)
726 p = &(*p)->rb_left;
Namhyung Kim6038def2011-05-24 17:11:22 -0700727 else if (vma->vm_start > pvma->vm_start) {
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->vm_end < pvma->vm_end)
David Howells8feae132009-01-08 12:04:47 +0000731 p = &(*p)->rb_left;
Namhyung Kim6038def2011-05-24 17:11:22 -0700732 else if (vma->vm_end > pvma->vm_end) {
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 if (vma < pvma)
David Howells8feae132009-01-08 12:04:47 +0000736 p = &(*p)->rb_left;
Namhyung Kim6038def2011-05-24 17:11:22 -0700737 else if (vma > pvma) {
738 rb_prev = parent;
David Howells8feae132009-01-08 12:04:47 +0000739 p = &(*p)->rb_right;
Namhyung Kim6038def2011-05-24 17:11:22 -0700740 } else
David Howells8feae132009-01-08 12:04:47 +0000741 BUG();
742 }
743
744 rb_link_node(&vma->vm_rb, parent, p);
745 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
746
747 /* add VMA to the VMA list also */
Namhyung Kim6038def2011-05-24 17:11:22 -0700748 prev = NULL;
749 if (rb_prev)
750 prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
David Howells30340972006-09-27 01:50:20 -0700751
Namhyung Kim6038def2011-05-24 17:11:22 -0700752 __vma_link_list(mm, vma, prev, parent);
David Howells8feae132009-01-08 12:04:47 +0000753}
754
755/*
756 * delete a VMA from its owning mm_struct and address space
757 */
758static void delete_vma_from_mm(struct vm_area_struct *vma)
759{
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700760 int i;
David Howells8feae132009-01-08 12:04:47 +0000761 struct address_space *mapping;
762 struct mm_struct *mm = vma->vm_mm;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700763 struct task_struct *curr = current;
David Howells8feae132009-01-08 12:04:47 +0000764
Bernd Schmidteb8cdec2009-09-21 17:03:57 -0700765 protect_vma(vma, 0);
766
David Howells8feae132009-01-08 12:04:47 +0000767 mm->map_count--;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700768 for (i = 0; i < VMACACHE_SIZE; i++) {
769 /* if the vma is cached, invalidate the entire cache */
770 if (curr->vmacache[i] == vma) {
Steven Miaoe020d5b2014-06-23 13:22:02 -0700771 vmacache_invalidate(mm);
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700772 break;
773 }
774 }
David Howells8feae132009-01-08 12:04:47 +0000775
776 /* remove the VMA from the mapping */
777 if (vma->vm_file) {
778 mapping = vma->vm_file->f_mapping;
779
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800780 i_mmap_lock_write(mapping);
David Howells8feae132009-01-08 12:04:47 +0000781 flush_dcache_mmap_lock(mapping);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700782 vma_interval_tree_remove(vma, &mapping->i_mmap);
David Howells8feae132009-01-08 12:04:47 +0000783 flush_dcache_mmap_unlock(mapping);
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800784 i_mmap_unlock_write(mapping);
David Howells8feae132009-01-08 12:04:47 +0000785 }
786
787 /* remove from the MM's tree and list */
788 rb_erase(&vma->vm_rb, &mm->mm_rb);
Namhyung Kimb951bf22011-05-24 17:11:23 -0700789
790 if (vma->vm_prev)
791 vma->vm_prev->vm_next = vma->vm_next;
792 else
793 mm->mmap = vma->vm_next;
794
795 if (vma->vm_next)
796 vma->vm_next->vm_prev = vma->vm_prev;
David Howells8feae132009-01-08 12:04:47 +0000797}
798
799/*
800 * destroy a VMA record
801 */
802static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
803{
David Howells8feae132009-01-08 12:04:47 +0000804 if (vma->vm_ops && vma->vm_ops->close)
805 vma->vm_ops->close(vma);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -0700806 if (vma->vm_file)
David Howells8feae132009-01-08 12:04:47 +0000807 fput(vma->vm_file);
David Howells8feae132009-01-08 12:04:47 +0000808 put_nommu_region(vma->vm_region);
809 kmem_cache_free(vm_area_cachep, vma);
David Howells30340972006-09-27 01:50:20 -0700810}
811
812/*
813 * look up the first VMA in which addr resides, NULL if none
814 * - should be called with mm->mmap_sem at least held readlocked
815 */
816struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
817{
David Howells8feae132009-01-08 12:04:47 +0000818 struct vm_area_struct *vma;
David Howells30340972006-09-27 01:50:20 -0700819
David Howells8feae132009-01-08 12:04:47 +0000820 /* check the cache first */
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700821 vma = vmacache_find(mm, addr);
822 if (likely(vma))
David Howells8feae132009-01-08 12:04:47 +0000823 return vma;
824
Namhyung Kime922c4c2011-05-24 17:11:24 -0700825 /* trawl the list (there may be multiple mappings in which addr
David Howells8feae132009-01-08 12:04:47 +0000826 * resides) */
Namhyung Kime922c4c2011-05-24 17:11:24 -0700827 for (vma = mm->mmap; vma; vma = vma->vm_next) {
David Howells8feae132009-01-08 12:04:47 +0000828 if (vma->vm_start > addr)
829 return NULL;
830 if (vma->vm_end > addr) {
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700831 vmacache_update(addr, vma);
David Howells8feae132009-01-08 12:04:47 +0000832 return vma;
833 }
David Howells30340972006-09-27 01:50:20 -0700834 }
835
David Howells30340972006-09-27 01:50:20 -0700836 return NULL;
837}
838EXPORT_SYMBOL(find_vma);
839
840/*
David Howells930e6522006-09-27 01:50:22 -0700841 * find a VMA
842 * - we don't extend stack VMAs under NOMMU conditions
843 */
844struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
845{
David Howells7561e8c2010-03-25 16:48:38 +0000846 return find_vma(mm, addr);
David Howells930e6522006-09-27 01:50:22 -0700847}
848
David Howells8feae132009-01-08 12:04:47 +0000849/*
850 * expand a stack to a given address
851 * - not supported under NOMMU conditions
852 */
Greg Ungerer57c8f632007-07-15 23:38:28 -0700853int expand_stack(struct vm_area_struct *vma, unsigned long address)
854{
855 return -ENOMEM;
856}
857
David Howells930e6522006-09-27 01:50:22 -0700858/*
David Howells6fa5f802006-09-27 01:50:21 -0700859 * look up the first VMA exactly that exactly matches addr
860 * - should be called with mm->mmap_sem at least held readlocked
861 */
David Howells8feae132009-01-08 12:04:47 +0000862static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
863 unsigned long addr,
864 unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865{
866 struct vm_area_struct *vma;
David Howells8feae132009-01-08 12:04:47 +0000867 unsigned long end = addr + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868
David Howells8feae132009-01-08 12:04:47 +0000869 /* check the cache first */
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700870 vma = vmacache_find_exact(mm, addr, end);
871 if (vma)
David Howells8feae132009-01-08 12:04:47 +0000872 return vma;
873
Namhyung Kime922c4c2011-05-24 17:11:24 -0700874 /* trawl the list (there may be multiple mappings in which addr
David Howells8feae132009-01-08 12:04:47 +0000875 * resides) */
Namhyung Kime922c4c2011-05-24 17:11:24 -0700876 for (vma = mm->mmap; vma; vma = vma->vm_next) {
David Howells8feae132009-01-08 12:04:47 +0000877 if (vma->vm_start < addr)
878 continue;
879 if (vma->vm_start > addr)
880 return NULL;
881 if (vma->vm_end == end) {
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700882 vmacache_update(addr, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 return vma;
David Howells8feae132009-01-08 12:04:47 +0000884 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 }
886
887 return NULL;
888}
889
David Howells30340972006-09-27 01:50:20 -0700890/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 * determine whether a mapping should be permitted and, if so, what sort of
892 * mapping we're capable of supporting
893 */
894static int validate_mmap_request(struct file *file,
895 unsigned long addr,
896 unsigned long len,
897 unsigned long prot,
898 unsigned long flags,
899 unsigned long pgoff,
900 unsigned long *_capabilities)
901{
David Howells8feae132009-01-08 12:04:47 +0000902 unsigned long capabilities, rlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 int ret;
904
905 /* do the simple checks first */
Leon Romanovsky22cc8772015-06-24 16:57:47 -0700906 if (flags & MAP_FIXED)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908
909 if ((flags & MAP_TYPE) != MAP_PRIVATE &&
910 (flags & MAP_TYPE) != MAP_SHARED)
911 return -EINVAL;
912
Mike Frysingerf81cff0d2006-12-06 12:02:59 +1000913 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 return -EINVAL;
915
Mike Frysingerf81cff0d2006-12-06 12:02:59 +1000916 /* Careful about overflows.. */
David Howells8feae132009-01-08 12:04:47 +0000917 rlen = PAGE_ALIGN(len);
918 if (!rlen || rlen > TASK_SIZE)
Mike Frysingerf81cff0d2006-12-06 12:02:59 +1000919 return -ENOMEM;
920
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 /* offset overflow? */
David Howells8feae132009-01-08 12:04:47 +0000922 if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
Mike Frysingerf81cff0d2006-12-06 12:02:59 +1000923 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924
925 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926 /* files must support mmap */
Al Viro72c2d532013-09-22 16:27:52 -0400927 if (!file->f_op->mmap)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928 return -ENODEV;
929
930 /* work out if what we've got could possibly be shared
931 * - we support chardevs that provide their own "memory"
932 * - we support files/blockdevs that are memory backed
933 */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100934 if (file->f_op->mmap_capabilities) {
935 capabilities = file->f_op->mmap_capabilities(file);
936 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 /* no explicit capabilities set, so assume some
938 * defaults */
Al Viro496ad9a2013-01-23 17:07:38 -0500939 switch (file_inode(file)->i_mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 case S_IFREG:
941 case S_IFBLK:
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100942 capabilities = NOMMU_MAP_COPY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943 break;
944
945 case S_IFCHR:
946 capabilities =
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100947 NOMMU_MAP_DIRECT |
948 NOMMU_MAP_READ |
949 NOMMU_MAP_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950 break;
951
952 default:
953 return -EINVAL;
954 }
955 }
956
957 /* eliminate any capabilities that we can't support on this
958 * device */
959 if (!file->f_op->get_unmapped_area)
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100960 capabilities &= ~NOMMU_MAP_DIRECT;
Al Viro6e242a12015-03-31 12:35:13 -0400961 if (!(file->f_mode & FMODE_CAN_READ))
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100962 capabilities &= ~NOMMU_MAP_COPY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963
Graff Yang28d7a6a2009-08-18 14:11:17 -0700964 /* The file shall have been opened with read permission. */
965 if (!(file->f_mode & FMODE_READ))
966 return -EACCES;
967
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 if (flags & MAP_SHARED) {
969 /* do checks for writing, appending and locking */
970 if ((prot & PROT_WRITE) &&
971 !(file->f_mode & FMODE_WRITE))
972 return -EACCES;
973
Al Viro496ad9a2013-01-23 17:07:38 -0500974 if (IS_APPEND(file_inode(file)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 (file->f_mode & FMODE_WRITE))
976 return -EACCES;
977
Jeff Laytond7a06982014-03-10 09:54:15 -0400978 if (locks_verify_locked(file))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 return -EAGAIN;
980
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100981 if (!(capabilities & NOMMU_MAP_DIRECT))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 return -ENODEV;
983
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 /* we mustn't privatise shared mappings */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100985 capabilities &= ~NOMMU_MAP_COPY;
Choi Gi-yongac714902014-04-07 15:37:36 -0700986 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 /* we're going to read the file into private memory we
988 * allocate */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100989 if (!(capabilities & NOMMU_MAP_COPY))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 return -ENODEV;
991
992 /* we don't permit a private writable mapping to be
993 * shared with the backing device */
994 if (prot & PROT_WRITE)
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100995 capabilities &= ~NOMMU_MAP_DIRECT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996 }
997
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100998 if (capabilities & NOMMU_MAP_DIRECT) {
999 if (((prot & PROT_READ) && !(capabilities & NOMMU_MAP_READ)) ||
1000 ((prot & PROT_WRITE) && !(capabilities & NOMMU_MAP_WRITE)) ||
1001 ((prot & PROT_EXEC) && !(capabilities & NOMMU_MAP_EXEC))
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001002 ) {
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001003 capabilities &= ~NOMMU_MAP_DIRECT;
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001004 if (flags & MAP_SHARED) {
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001005 pr_warn("MAP_SHARED not completely supported on !MMU\n");
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001006 return -EINVAL;
1007 }
1008 }
1009 }
1010
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 /* handle executable mappings and implied executable
1012 * mappings */
Eric W. Biederman90f85722015-06-29 14:42:03 -05001013 if (path_noexec(&file->f_path)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 if (prot & PROT_EXEC)
1015 return -EPERM;
Choi Gi-yongac714902014-04-07 15:37:36 -07001016 } else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 /* handle implication of PROT_EXEC by PROT_READ */
1018 if (current->personality & READ_IMPLIES_EXEC) {
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001019 if (capabilities & NOMMU_MAP_EXEC)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 prot |= PROT_EXEC;
1021 }
Choi Gi-yongac714902014-04-07 15:37:36 -07001022 } else if ((prot & PROT_READ) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 (prot & PROT_EXEC) &&
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001024 !(capabilities & NOMMU_MAP_EXEC)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 ) {
1026 /* backing file is not executable, try to copy */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001027 capabilities &= ~NOMMU_MAP_DIRECT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 }
Choi Gi-yongac714902014-04-07 15:37:36 -07001029 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 /* anonymous mappings are always memory backed and can be
1031 * privately mapped
1032 */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001033 capabilities = NOMMU_MAP_COPY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034
1035 /* handle PROT_EXEC implication by PROT_READ */
1036 if ((prot & PROT_READ) &&
1037 (current->personality & READ_IMPLIES_EXEC))
1038 prot |= PROT_EXEC;
1039 }
1040
1041 /* allow the security API to have its say */
Al Viroe5467852012-05-30 13:30:51 -04001042 ret = security_mmap_addr(addr);
1043 if (ret < 0)
1044 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045
1046 /* looks okay */
1047 *_capabilities = capabilities;
1048 return 0;
1049}
1050
1051/*
1052 * we've determined that we can make the mapping, now translate what we
1053 * now know into VMA flags
1054 */
1055static unsigned long determine_vm_flags(struct file *file,
1056 unsigned long prot,
1057 unsigned long flags,
1058 unsigned long capabilities)
1059{
1060 unsigned long vm_flags;
1061
Dave Hansene6bfb702016-02-12 13:02:31 -08001062 vm_flags = calc_vm_prot_bits(prot, 0) | calc_vm_flag_bits(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 /* vm_flags |= mm->def_flags; */
1064
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001065 if (!(capabilities & NOMMU_MAP_DIRECT)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 /* attempt to share read-only copies of mapped file chunks */
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001067 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 if (file && !(prot & PROT_WRITE))
1069 vm_flags |= VM_MAYSHARE;
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001070 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071 /* overlay a shareable mapping on the backing device or inode
1072 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1073 * romfs/cramfs */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001074 vm_flags |= VM_MAYSHARE | (capabilities & NOMMU_VMFLAGS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 if (flags & MAP_SHARED)
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001076 vm_flags |= VM_SHARED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077 }
1078
1079 /* refuse to let anyone share private mappings with this process if
1080 * it's being traced - otherwise breakpoints set in it may interfere
1081 * with another untraced process
1082 */
Tejun Heoa288eec2011-06-17 16:50:37 +02001083 if ((flags & MAP_PRIVATE) && current->ptrace)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 vm_flags &= ~VM_MAYSHARE;
1085
1086 return vm_flags;
1087}
1088
1089/*
David Howells8feae132009-01-08 12:04:47 +00001090 * set up a shared mapping on a file (the driver or filesystem provides and
1091 * pins the storage)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 */
David Howells8feae132009-01-08 12:04:47 +00001093static int do_mmap_shared_file(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094{
1095 int ret;
1096
1097 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001098 if (ret == 0) {
1099 vma->vm_region->vm_top = vma->vm_region->vm_end;
David Howells645d83c2009-09-24 15:13:10 +01001100 return 0;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001101 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 if (ret != -ENOSYS)
1103 return ret;
1104
David Howells3fa30462010-03-23 13:35:21 -07001105 /* getting -ENOSYS indicates that direct mmap isn't possible (as
1106 * opposed to tried but failed) so we can only give a suitable error as
1107 * it's not possible to make a private copy if MAP_SHARED was given */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 return -ENODEV;
1109}
1110
1111/*
1112 * set up a private mapping or an anonymous shared mapping
1113 */
David Howells8feae132009-01-08 12:04:47 +00001114static int do_mmap_private(struct vm_area_struct *vma,
1115 struct vm_region *region,
David Howells645d83c2009-09-24 15:13:10 +01001116 unsigned long len,
1117 unsigned long capabilities)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118{
Joonsoo Kimdbc83582014-12-12 16:55:55 -08001119 unsigned long total, point;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 void *base;
David Howells8feae132009-01-08 12:04:47 +00001121 int ret, order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122
1123 /* invoke the file's mapping function so that it can keep track of
1124 * shared mappings on devices or memory
1125 * - VM_MAYSHARE will be set if it may attempt to share
1126 */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001127 if (capabilities & NOMMU_MAP_DIRECT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001129 if (ret == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 /* shouldn't return success if we're not sharing */
Paul Mundtdd8632a2009-01-08 12:04:47 +00001131 BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1132 vma->vm_region->vm_top = vma->vm_region->vm_end;
David Howells645d83c2009-09-24 15:13:10 +01001133 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134 }
Paul Mundtdd8632a2009-01-08 12:04:47 +00001135 if (ret != -ENOSYS)
1136 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137
1138 /* getting an ENOSYS error indicates that direct mmap isn't
1139 * possible (as opposed to tried but failed) so we'll try to
1140 * make a private copy of the data and map that instead */
1141 }
1142
David Howells8feae132009-01-08 12:04:47 +00001143
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 /* allocate some memory to hold the mapping
1145 * - note that this may not return a page-aligned address if the object
1146 * we're allocating is smaller than a page
1147 */
Bob Liuf67d9b12011-05-24 17:12:56 -07001148 order = get_order(len);
David Howells8feae132009-01-08 12:04:47 +00001149 total = 1 << order;
Bob Liuf67d9b12011-05-24 17:12:56 -07001150 point = len >> PAGE_SHIFT;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001151
Joonsoo Kimdbc83582014-12-12 16:55:55 -08001152 /* we don't want to allocate a power-of-2 sized page set */
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001153 if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages)
Joonsoo Kimdbc83582014-12-12 16:55:55 -08001154 total = point;
David Howells8feae132009-01-08 12:04:47 +00001155
Joonsoo Kimda616532015-02-27 15:51:43 -08001156 base = alloc_pages_exact(total << PAGE_SHIFT, GFP_KERNEL);
Joonsoo Kimdbc83582014-12-12 16:55:55 -08001157 if (!base)
1158 goto enomem;
David Howells8feae132009-01-08 12:04:47 +00001159
Joonsoo Kimdbc83582014-12-12 16:55:55 -08001160 atomic_long_add(total, &mmap_pages_allocated);
1161
David Howells8feae132009-01-08 12:04:47 +00001162 region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1163 region->vm_start = (unsigned long) base;
Bob Liuf67d9b12011-05-24 17:12:56 -07001164 region->vm_end = region->vm_start + len;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001165 region->vm_top = region->vm_start + (total << PAGE_SHIFT);
David Howells8feae132009-01-08 12:04:47 +00001166
1167 vma->vm_start = region->vm_start;
1168 vma->vm_end = region->vm_start + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169
1170 if (vma->vm_file) {
1171 /* read the contents of a file into the copy */
1172 mm_segment_t old_fs;
1173 loff_t fpos;
1174
1175 fpos = vma->vm_pgoff;
1176 fpos <<= PAGE_SHIFT;
1177
1178 old_fs = get_fs();
1179 set_fs(KERNEL_DS);
Al Viro6e242a12015-03-31 12:35:13 -04001180 ret = __vfs_read(vma->vm_file, base, len, &fpos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181 set_fs(old_fs);
1182
1183 if (ret < 0)
1184 goto error_free;
1185
1186 /* clear the last little bit */
Bob Liuf67d9b12011-05-24 17:12:56 -07001187 if (ret < len)
1188 memset(base + ret, 0, len - ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 }
1191
1192 return 0;
1193
1194error_free:
Namhyung Kim7223bb42011-05-24 17:11:26 -07001195 free_page_series(region->vm_start, region->vm_top);
David Howells8feae132009-01-08 12:04:47 +00001196 region->vm_start = vma->vm_start = 0;
1197 region->vm_end = vma->vm_end = 0;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001198 region->vm_top = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 return ret;
1200
1201enomem:
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -07001202 pr_err("Allocation of length %lu from process %d (%s) failed\n",
Greg Ungerer05ae6fa2009-01-13 17:30:22 +10001203 len, current->pid, current->comm);
David Rientjes7bf02ea2011-05-24 17:11:16 -07001204 show_free_areas(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 return -ENOMEM;
1206}
1207
1208/*
1209 * handle mapping creation for uClinux
1210 */
Oleg Nesterov1fcfd8d2015-09-09 15:39:29 -07001211unsigned long do_mmap(struct file *file,
1212 unsigned long addr,
1213 unsigned long len,
1214 unsigned long prot,
1215 unsigned long flags,
1216 vm_flags_t vm_flags,
1217 unsigned long pgoff,
1218 unsigned long *populate)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219{
David Howells8feae132009-01-08 12:04:47 +00001220 struct vm_area_struct *vma;
1221 struct vm_region *region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 struct rb_node *rb;
Oleg Nesterov1fcfd8d2015-09-09 15:39:29 -07001223 unsigned long capabilities, result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 int ret;
1225
Michel Lespinasse41badc12013-02-22 16:32:47 -08001226 *populate = 0;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001227
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 /* decide whether we should attempt the mapping, and if so what sort of
1229 * mapping */
1230 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1231 &capabilities);
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001232 if (ret < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233 return ret;
1234
David Howells06aab5a2009-09-24 12:33:48 +01001235 /* we ignore the address hint */
1236 addr = 0;
Bob Liuf67d9b12011-05-24 17:12:56 -07001237 len = PAGE_ALIGN(len);
David Howells06aab5a2009-09-24 12:33:48 +01001238
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 /* we've determined that we can make the mapping, now translate what we
1240 * now know into VMA flags */
Oleg Nesterov1fcfd8d2015-09-09 15:39:29 -07001241 vm_flags |= determine_vm_flags(file, prot, flags, capabilities);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242
David Howells8feae132009-01-08 12:04:47 +00001243 /* we're going to need to record the mapping */
1244 region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1245 if (!region)
1246 goto error_getting_region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247
David Howells8feae132009-01-08 12:04:47 +00001248 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1249 if (!vma)
1250 goto error_getting_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251
David Howells1e2ae592010-01-15 17:01:33 -08001252 region->vm_usage = 1;
David Howells8feae132009-01-08 12:04:47 +00001253 region->vm_flags = vm_flags;
1254 region->vm_pgoff = pgoff;
1255
Rik van Riel5beb4932010-03-05 13:42:07 -08001256 INIT_LIST_HEAD(&vma->anon_vma_chain);
David Howells8feae132009-01-08 12:04:47 +00001257 vma->vm_flags = vm_flags;
1258 vma->vm_pgoff = pgoff;
1259
1260 if (file) {
Al Virocb0942b2012-08-27 14:48:26 -04001261 region->vm_file = get_file(file);
1262 vma->vm_file = get_file(file);
David Howells8feae132009-01-08 12:04:47 +00001263 }
1264
1265 down_write(&nommu_region_sem);
1266
1267 /* if we want to share, we need to check for regions created by other
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 * mmap() calls that overlap with our proposed mapping
David Howells8feae132009-01-08 12:04:47 +00001269 * - we can only share with a superset match on most regular files
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270 * - shared mappings on character devices and memory backed files are
1271 * permitted to overlap inexactly as far as we are concerned for in
1272 * these cases, sharing is handled in the driver or filesystem rather
1273 * than here
1274 */
1275 if (vm_flags & VM_MAYSHARE) {
David Howells8feae132009-01-08 12:04:47 +00001276 struct vm_region *pregion;
1277 unsigned long pglen, rpglen, pgend, rpgend, start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278
David Howells8feae132009-01-08 12:04:47 +00001279 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1280 pgend = pgoff + pglen;
David Howells165b2392007-03-22 00:11:24 -08001281
David Howells8feae132009-01-08 12:04:47 +00001282 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1283 pregion = rb_entry(rb, struct vm_region, vm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284
David Howells8feae132009-01-08 12:04:47 +00001285 if (!(pregion->vm_flags & VM_MAYSHARE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 continue;
1287
1288 /* search for overlapping mappings on the same file */
Al Viro496ad9a2013-01-23 17:07:38 -05001289 if (file_inode(pregion->vm_file) !=
1290 file_inode(file))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291 continue;
1292
David Howells8feae132009-01-08 12:04:47 +00001293 if (pregion->vm_pgoff >= pgend)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294 continue;
1295
David Howells8feae132009-01-08 12:04:47 +00001296 rpglen = pregion->vm_end - pregion->vm_start;
1297 rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1298 rpgend = pregion->vm_pgoff + rpglen;
1299 if (pgoff >= rpgend)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300 continue;
1301
David Howells8feae132009-01-08 12:04:47 +00001302 /* handle inexactly overlapping matches between
1303 * mappings */
1304 if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1305 !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1306 /* new mapping is not a subset of the region */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001307 if (!(capabilities & NOMMU_MAP_DIRECT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308 goto sharing_violation;
1309 continue;
1310 }
1311
David Howells8feae132009-01-08 12:04:47 +00001312 /* we've found a region we can share */
David Howells1e2ae592010-01-15 17:01:33 -08001313 pregion->vm_usage++;
David Howells8feae132009-01-08 12:04:47 +00001314 vma->vm_region = pregion;
1315 start = pregion->vm_start;
1316 start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1317 vma->vm_start = start;
1318 vma->vm_end = start + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001320 if (pregion->vm_flags & VM_MAPPED_COPY)
David Howells8feae132009-01-08 12:04:47 +00001321 vma->vm_flags |= VM_MAPPED_COPY;
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001322 else {
David Howells8feae132009-01-08 12:04:47 +00001323 ret = do_mmap_shared_file(vma);
1324 if (ret < 0) {
1325 vma->vm_region = NULL;
1326 vma->vm_start = 0;
1327 vma->vm_end = 0;
David Howells1e2ae592010-01-15 17:01:33 -08001328 pregion->vm_usage--;
David Howells8feae132009-01-08 12:04:47 +00001329 pregion = NULL;
1330 goto error_just_free;
1331 }
1332 }
1333 fput(region->vm_file);
1334 kmem_cache_free(vm_region_jar, region);
1335 region = pregion;
1336 result = start;
1337 goto share;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 }
1339
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 /* obtain the address at which to make a shared mapping
1341 * - this is the hook for quasi-memory character devices to
1342 * tell us the location of a shared mapping
1343 */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001344 if (capabilities & NOMMU_MAP_DIRECT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345 addr = file->f_op->get_unmapped_area(file, addr, len,
1346 pgoff, flags);
Namhyung Kimbb005a52011-05-24 17:11:27 -07001347 if (IS_ERR_VALUE(addr)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 ret = addr;
Namhyung Kimbb005a52011-05-24 17:11:27 -07001349 if (ret != -ENOSYS)
David Howells8feae132009-01-08 12:04:47 +00001350 goto error_just_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351
1352 /* the driver refused to tell us where to site
1353 * the mapping so we'll have to attempt to copy
1354 * it */
Namhyung Kimbb005a52011-05-24 17:11:27 -07001355 ret = -ENODEV;
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001356 if (!(capabilities & NOMMU_MAP_COPY))
David Howells8feae132009-01-08 12:04:47 +00001357 goto error_just_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001359 capabilities &= ~NOMMU_MAP_DIRECT;
David Howells8feae132009-01-08 12:04:47 +00001360 } else {
1361 vma->vm_start = region->vm_start = addr;
1362 vma->vm_end = region->vm_end = addr + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 }
1364 }
1365 }
1366
David Howells8feae132009-01-08 12:04:47 +00001367 vma->vm_region = region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368
David Howells645d83c2009-09-24 15:13:10 +01001369 /* set up the mapping
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001370 * - the region is filled in if NOMMU_MAP_DIRECT is still set
David Howells645d83c2009-09-24 15:13:10 +01001371 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 if (file && vma->vm_flags & VM_SHARED)
David Howells8feae132009-01-08 12:04:47 +00001373 ret = do_mmap_shared_file(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 else
David Howells645d83c2009-09-24 15:13:10 +01001375 ret = do_mmap_private(vma, region, len, capabilities);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 if (ret < 0)
David Howells645d83c2009-09-24 15:13:10 +01001377 goto error_just_free;
1378 add_nommu_region(region);
David Howells8feae132009-01-08 12:04:47 +00001379
Jie Zhangea637632009-12-14 18:00:02 -08001380 /* clear anonymous mappings that don't ask for uninitialized data */
1381 if (!vma->vm_file && !(flags & MAP_UNINITIALIZED))
1382 memset((void *)region->vm_start, 0,
1383 region->vm_end - region->vm_start);
1384
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 /* okay... we have a mapping; now we have to register it */
David Howells8feae132009-01-08 12:04:47 +00001386 result = vma->vm_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 current->mm->total_vm += len >> PAGE_SHIFT;
1389
David Howells8feae132009-01-08 12:04:47 +00001390share:
1391 add_vma_to_mm(current->mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392
Mike Frysingercfe79c02010-01-06 17:23:23 +00001393 /* we flush the region from the icache only when the first executable
1394 * mapping of it is made */
1395 if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1396 flush_icache_range(region->vm_start, region->vm_end);
1397 region->vm_icache_flushed = true;
1398 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399
Mike Frysingercfe79c02010-01-06 17:23:23 +00001400 up_write(&nommu_region_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401
David Howells8feae132009-01-08 12:04:47 +00001402 return result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403
David Howells8feae132009-01-08 12:04:47 +00001404error_just_free:
1405 up_write(&nommu_region_sem);
1406error:
David Howells89a86402009-10-30 13:13:26 +00001407 if (region->vm_file)
1408 fput(region->vm_file);
David Howells8feae132009-01-08 12:04:47 +00001409 kmem_cache_free(vm_region_jar, region);
David Howells89a86402009-10-30 13:13:26 +00001410 if (vma->vm_file)
1411 fput(vma->vm_file);
David Howells8feae132009-01-08 12:04:47 +00001412 kmem_cache_free(vm_area_cachep, vma);
David Howells8feae132009-01-08 12:04:47 +00001413 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414
David Howells8feae132009-01-08 12:04:47 +00001415sharing_violation:
1416 up_write(&nommu_region_sem);
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001417 pr_warn("Attempt to share mismatched mappings\n");
David Howells8feae132009-01-08 12:04:47 +00001418 ret = -EINVAL;
1419 goto error;
1420
1421error_getting_vma:
1422 kmem_cache_free(vm_region_jar, region);
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001423 pr_warn("Allocation of vma 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
David Howells8feae132009-01-08 12:04:47 +00001428error_getting_region:
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001429 pr_warn("Allocation of vm region for %lu byte allocation from process %d failed\n",
1430 len, current->pid);
David Rientjes7bf02ea2011-05-24 17:11:16 -07001431 show_free_areas(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432 return -ENOMEM;
1433}
Linus Torvalds6be5ceb2012-04-20 17:13:58 -07001434
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001435SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1436 unsigned long, prot, unsigned long, flags,
1437 unsigned long, fd, unsigned long, pgoff)
1438{
1439 struct file *file = NULL;
1440 unsigned long retval = -EBADF;
1441
Al Viro120a7952010-10-30 02:54:44 -04001442 audit_mmap_fd(fd, flags);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001443 if (!(flags & MAP_ANONYMOUS)) {
1444 file = fget(fd);
1445 if (!file)
1446 goto out;
1447 }
1448
1449 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1450
Greg Ungererad1ed292012-06-04 14:29:59 +10001451 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001452
1453 if (file)
1454 fput(file);
1455out:
1456 return retval;
1457}
1458
Christoph Hellwiga4679372010-03-10 15:21:15 -08001459#ifdef __ARCH_WANT_SYS_OLD_MMAP
1460struct mmap_arg_struct {
1461 unsigned long addr;
1462 unsigned long len;
1463 unsigned long prot;
1464 unsigned long flags;
1465 unsigned long fd;
1466 unsigned long offset;
1467};
1468
1469SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1470{
1471 struct mmap_arg_struct a;
1472
1473 if (copy_from_user(&a, arg, sizeof(a)))
1474 return -EFAULT;
Alexander Kuleshov1824cb72015-11-05 18:46:35 -08001475 if (offset_in_page(a.offset))
Christoph Hellwiga4679372010-03-10 15:21:15 -08001476 return -EINVAL;
1477
1478 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1479 a.offset >> PAGE_SHIFT);
1480}
1481#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1482
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483/*
David Howells8feae132009-01-08 12:04:47 +00001484 * split a vma into two pieces at address 'addr', a new vma is allocated either
1485 * for the first part or the tail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 */
David Howells8feae132009-01-08 12:04:47 +00001487int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1488 unsigned long addr, int new_below)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489{
David Howells8feae132009-01-08 12:04:47 +00001490 struct vm_area_struct *new;
1491 struct vm_region *region;
1492 unsigned long npages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493
David Howells779c1022010-01-15 17:01:34 -08001494 /* we're only permitted to split anonymous regions (these should have
1495 * only a single usage on the region) */
1496 if (vma->vm_file)
David Howells8feae132009-01-08 12:04:47 +00001497 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498
David Howells8feae132009-01-08 12:04:47 +00001499 if (mm->map_count >= sysctl_max_map_count)
1500 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501
David Howells8feae132009-01-08 12:04:47 +00001502 region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1503 if (!region)
1504 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505
David Howells8feae132009-01-08 12:04:47 +00001506 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1507 if (!new) {
1508 kmem_cache_free(vm_region_jar, region);
1509 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 }
David Howells8feae132009-01-08 12:04:47 +00001511
1512 /* most fields are the same, copy all, and then fixup */
1513 *new = *vma;
1514 *region = *vma->vm_region;
1515 new->vm_region = region;
1516
1517 npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1518
1519 if (new_below) {
Paul Mundtdd8632a2009-01-08 12:04:47 +00001520 region->vm_top = region->vm_end = new->vm_end = addr;
David Howells8feae132009-01-08 12:04:47 +00001521 } else {
1522 region->vm_start = new->vm_start = addr;
1523 region->vm_pgoff = new->vm_pgoff += npages;
1524 }
1525
1526 if (new->vm_ops && new->vm_ops->open)
1527 new->vm_ops->open(new);
1528
1529 delete_vma_from_mm(vma);
1530 down_write(&nommu_region_sem);
1531 delete_nommu_region(vma->vm_region);
1532 if (new_below) {
1533 vma->vm_region->vm_start = vma->vm_start = addr;
1534 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1535 } else {
1536 vma->vm_region->vm_end = vma->vm_end = addr;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001537 vma->vm_region->vm_top = addr;
David Howells8feae132009-01-08 12:04:47 +00001538 }
1539 add_nommu_region(vma->vm_region);
1540 add_nommu_region(new->vm_region);
1541 up_write(&nommu_region_sem);
1542 add_vma_to_mm(mm, vma);
1543 add_vma_to_mm(mm, new);
1544 return 0;
1545}
1546
1547/*
1548 * shrink a VMA by removing the specified chunk from either the beginning or
1549 * the end
1550 */
1551static int shrink_vma(struct mm_struct *mm,
1552 struct vm_area_struct *vma,
1553 unsigned long from, unsigned long to)
1554{
1555 struct vm_region *region;
1556
David Howells8feae132009-01-08 12:04:47 +00001557 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1558 * and list */
1559 delete_vma_from_mm(vma);
1560 if (from > vma->vm_start)
1561 vma->vm_end = from;
1562 else
1563 vma->vm_start = to;
1564 add_vma_to_mm(mm, vma);
1565
1566 /* cut the backing region down to size */
1567 region = vma->vm_region;
David Howells1e2ae592010-01-15 17:01:33 -08001568 BUG_ON(region->vm_usage != 1);
David Howells8feae132009-01-08 12:04:47 +00001569
1570 down_write(&nommu_region_sem);
1571 delete_nommu_region(region);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001572 if (from > region->vm_start) {
1573 to = region->vm_top;
1574 region->vm_top = region->vm_end = from;
1575 } else {
David Howells8feae132009-01-08 12:04:47 +00001576 region->vm_start = to;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001577 }
David Howells8feae132009-01-08 12:04:47 +00001578 add_nommu_region(region);
1579 up_write(&nommu_region_sem);
1580
1581 free_page_series(from, to);
1582 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583}
1584
David Howells30340972006-09-27 01:50:20 -07001585/*
1586 * release a mapping
David Howells8feae132009-01-08 12:04:47 +00001587 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1588 * VMA, though it need not cover the whole VMA
David Howells30340972006-09-27 01:50:20 -07001589 */
David Howells8feae132009-01-08 12:04:47 +00001590int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591{
David Howells8feae132009-01-08 12:04:47 +00001592 struct vm_area_struct *vma;
Bob Liuf67d9b12011-05-24 17:12:56 -07001593 unsigned long end;
David Howells8feae132009-01-08 12:04:47 +00001594 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595
Bob Liuf67d9b12011-05-24 17:12:56 -07001596 len = PAGE_ALIGN(len);
David Howells8feae132009-01-08 12:04:47 +00001597 if (len == 0)
1598 return -EINVAL;
1599
Bob Liuf67d9b12011-05-24 17:12:56 -07001600 end = start + len;
1601
David Howells8feae132009-01-08 12:04:47 +00001602 /* find the first potentially overlapping VMA */
1603 vma = find_vma(mm, start);
1604 if (!vma) {
Choi Gi-yongac714902014-04-07 15:37:36 -07001605 static int limit;
David Howells33e5d7692009-04-02 16:56:32 -07001606 if (limit < 5) {
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001607 pr_warn("munmap of memory not mmapped by process %d (%s): 0x%lx-0x%lx\n",
1608 current->pid, current->comm,
1609 start, start + len - 1);
David Howells33e5d7692009-04-02 16:56:32 -07001610 limit++;
1611 }
David Howells8feae132009-01-08 12:04:47 +00001612 return -EINVAL;
David Howells30340972006-09-27 01:50:20 -07001613 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614
David Howells8feae132009-01-08 12:04:47 +00001615 /* we're allowed to split an anonymous VMA but not a file-backed one */
1616 if (vma->vm_file) {
1617 do {
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001618 if (start > vma->vm_start)
David Howells8feae132009-01-08 12:04:47 +00001619 return -EINVAL;
David Howells8feae132009-01-08 12:04:47 +00001620 if (end == vma->vm_end)
1621 goto erase_whole_vma;
Namhyung Kimd75a3102011-05-24 17:11:25 -07001622 vma = vma->vm_next;
1623 } while (vma);
David Howells8feae132009-01-08 12:04:47 +00001624 return -EINVAL;
1625 } else {
1626 /* the chunk must be a subset of the VMA found */
1627 if (start == vma->vm_start && end == vma->vm_end)
1628 goto erase_whole_vma;
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001629 if (start < vma->vm_start || end > vma->vm_end)
David Howells8feae132009-01-08 12:04:47 +00001630 return -EINVAL;
Alexander Kuleshov1824cb72015-11-05 18:46:35 -08001631 if (offset_in_page(start))
David Howells8feae132009-01-08 12:04:47 +00001632 return -EINVAL;
Alexander Kuleshov1824cb72015-11-05 18:46:35 -08001633 if (end != vma->vm_end && offset_in_page(end))
David Howells8feae132009-01-08 12:04:47 +00001634 return -EINVAL;
David Howells8feae132009-01-08 12:04:47 +00001635 if (start != vma->vm_start && end != vma->vm_end) {
1636 ret = split_vma(mm, vma, start, 1);
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001637 if (ret < 0)
David Howells8feae132009-01-08 12:04:47 +00001638 return ret;
David Howells8feae132009-01-08 12:04:47 +00001639 }
1640 return shrink_vma(mm, vma, start, end);
1641 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642
David Howells8feae132009-01-08 12:04:47 +00001643erase_whole_vma:
1644 delete_vma_from_mm(vma);
1645 delete_vma(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646 return 0;
1647}
Paul Mundtb5073172007-07-21 04:37:25 -07001648EXPORT_SYMBOL(do_munmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649
Al Virobfce2812012-04-20 21:57:04 -04001650int vm_munmap(unsigned long addr, size_t len)
David Howells30340972006-09-27 01:50:20 -07001651{
Al Virobfce2812012-04-20 21:57:04 -04001652 struct mm_struct *mm = current->mm;
David Howells30340972006-09-27 01:50:20 -07001653 int ret;
David Howells30340972006-09-27 01:50:20 -07001654
1655 down_write(&mm->mmap_sem);
1656 ret = do_munmap(mm, addr, len);
1657 up_write(&mm->mmap_sem);
1658 return ret;
1659}
Linus Torvaldsa46ef992012-04-20 16:20:01 -07001660EXPORT_SYMBOL(vm_munmap);
1661
1662SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1663{
Al Virobfce2812012-04-20 21:57:04 -04001664 return vm_munmap(addr, len);
Linus Torvaldsa46ef992012-04-20 16:20:01 -07001665}
David Howells30340972006-09-27 01:50:20 -07001666
1667/*
David Howells8feae132009-01-08 12:04:47 +00001668 * release all the mappings made in a process's VM space
David Howells30340972006-09-27 01:50:20 -07001669 */
David Howells8feae132009-01-08 12:04:47 +00001670void exit_mmap(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671{
David Howells8feae132009-01-08 12:04:47 +00001672 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673
David Howells8feae132009-01-08 12:04:47 +00001674 if (!mm)
1675 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676
David Howells8feae132009-01-08 12:04:47 +00001677 mm->total_vm = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678
David Howells8feae132009-01-08 12:04:47 +00001679 while ((vma = mm->mmap)) {
1680 mm->mmap = vma->vm_next;
1681 delete_vma_from_mm(vma);
1682 delete_vma(mm, vma);
Steven J. Magnani04c34962010-11-24 12:56:54 -08001683 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684 }
1685}
1686
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07001687unsigned long vm_brk(unsigned long addr, unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688{
1689 return -ENOMEM;
1690}
1691
1692/*
David Howells6fa5f802006-09-27 01:50:21 -07001693 * expand (or shrink) an existing mapping, potentially moving it at the same
1694 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695 *
David Howells6fa5f802006-09-27 01:50:21 -07001696 * under NOMMU conditions, we only permit changing a mapping's size, and only
David Howells8feae132009-01-08 12:04:47 +00001697 * as long as it stays within the region allocated by do_mmap_private() and the
1698 * block is not shareable
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 *
David Howells6fa5f802006-09-27 01:50:21 -07001700 * MREMAP_FIXED is not supported under NOMMU conditions
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701 */
Al Viro4b377ba2013-03-04 10:47:59 -05001702static unsigned long do_mremap(unsigned long addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703 unsigned long old_len, unsigned long new_len,
1704 unsigned long flags, unsigned long new_addr)
1705{
David Howells6fa5f802006-09-27 01:50:21 -07001706 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707
1708 /* insanity checks first */
Bob Liuf67d9b12011-05-24 17:12:56 -07001709 old_len = PAGE_ALIGN(old_len);
1710 new_len = PAGE_ALIGN(new_len);
David Howells8feae132009-01-08 12:04:47 +00001711 if (old_len == 0 || new_len == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712 return (unsigned long) -EINVAL;
1713
Alexander Kuleshov1824cb72015-11-05 18:46:35 -08001714 if (offset_in_page(addr))
David Howells8feae132009-01-08 12:04:47 +00001715 return -EINVAL;
1716
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717 if (flags & MREMAP_FIXED && new_addr != addr)
1718 return (unsigned long) -EINVAL;
1719
David Howells8feae132009-01-08 12:04:47 +00001720 vma = find_vma_exact(current->mm, addr, old_len);
David Howells6fa5f802006-09-27 01:50:21 -07001721 if (!vma)
1722 return (unsigned long) -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723
David Howells6fa5f802006-09-27 01:50:21 -07001724 if (vma->vm_end != vma->vm_start + old_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 return (unsigned long) -EFAULT;
1726
David Howells6fa5f802006-09-27 01:50:21 -07001727 if (vma->vm_flags & VM_MAYSHARE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728 return (unsigned long) -EPERM;
1729
David Howells8feae132009-01-08 12:04:47 +00001730 if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731 return (unsigned long) -ENOMEM;
1732
1733 /* all checks complete - do it */
David Howells6fa5f802006-09-27 01:50:21 -07001734 vma->vm_end = vma->vm_start + new_len;
David Howells6fa5f802006-09-27 01:50:21 -07001735 return vma->vm_start;
1736}
1737
Heiko Carstens6a6160a2009-01-14 14:14:15 +01001738SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1739 unsigned long, new_len, unsigned long, flags,
1740 unsigned long, new_addr)
David Howells6fa5f802006-09-27 01:50:21 -07001741{
1742 unsigned long ret;
1743
1744 down_write(&current->mm->mmap_sem);
1745 ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1746 up_write(&current->mm->mmap_sem);
1747 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748}
1749
Michel Lespinasse240aade2013-02-22 16:35:56 -08001750struct page *follow_page_mask(struct vm_area_struct *vma,
1751 unsigned long address, unsigned int flags,
1752 unsigned int *page_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753{
Michel Lespinasse240aade2013-02-22 16:35:56 -08001754 *page_mask = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 return NULL;
1756}
1757
Bob Liu8f3b1322011-07-08 15:39:46 -07001758int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1759 unsigned long pfn, unsigned long size, pgprot_t prot)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760{
Bob Liu8f3b1322011-07-08 15:39:46 -07001761 if (addr != (pfn << PAGE_SHIFT))
1762 return -EINVAL;
1763
Konstantin Khlebnikov314e51b2012-10-08 16:29:02 -07001764 vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
Greg Ungerer66aa2b42005-09-12 11:18:10 +10001765 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766}
Luke Yang22c4af42006-07-14 00:24:09 -07001767EXPORT_SYMBOL(remap_pfn_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768
Linus Torvalds3c0b9de2013-04-27 13:25:38 -07001769int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)
1770{
1771 unsigned long pfn = start >> PAGE_SHIFT;
1772 unsigned long vm_len = vma->vm_end - vma->vm_start;
1773
1774 pfn += vma->vm_pgoff;
1775 return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);
1776}
1777EXPORT_SYMBOL(vm_iomap_memory);
1778
Paul Mundtf905bc42008-02-04 22:29:59 -08001779int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1780 unsigned long pgoff)
1781{
1782 unsigned int size = vma->vm_end - vma->vm_start;
1783
1784 if (!(vma->vm_flags & VM_USERMAP))
1785 return -EINVAL;
1786
1787 vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1788 vma->vm_end = vma->vm_start + size;
1789
1790 return 0;
1791}
1792EXPORT_SYMBOL(remap_vmalloc_range);
1793
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1795 unsigned long len, unsigned long pgoff, unsigned long flags)
1796{
1797 return -ENOMEM;
1798}
1799
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800void unmap_mapping_range(struct address_space *mapping,
1801 loff_t const holebegin, loff_t const holelen,
1802 int even_cows)
1803{
1804}
Luke Yang22c4af42006-07-14 00:24:09 -07001805EXPORT_SYMBOL(unmap_mapping_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806
Nick Piggind0217ac2007-07-19 01:47:03 -07001807int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
David Howellsb0e15192006-01-06 00:11:42 -08001808{
1809 BUG();
Nick Piggind0217ac2007-07-19 01:47:03 -07001810 return 0;
David Howellsb0e15192006-01-06 00:11:42 -08001811}
Paul Mundtb5073172007-07-21 04:37:25 -07001812EXPORT_SYMBOL(filemap_fault);
David Howells0ec76a12006-09-27 01:50:15 -07001813
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07001814void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf)
1815{
1816 BUG();
1817}
1818EXPORT_SYMBOL(filemap_map_pages);
1819
Mike Frysingerf55f1992011-03-29 14:05:12 +01001820static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
1821 unsigned long addr, void *buf, int len, int write)
David Howells0ec76a12006-09-27 01:50:15 -07001822{
David Howells0ec76a12006-09-27 01:50:15 -07001823 struct vm_area_struct *vma;
David Howells0ec76a12006-09-27 01:50:15 -07001824
1825 down_read(&mm->mmap_sem);
1826
1827 /* the access must start within one of the target process's mappings */
David Howells0159b142006-09-27 01:50:16 -07001828 vma = find_vma(mm, addr);
1829 if (vma) {
David Howells0ec76a12006-09-27 01:50:15 -07001830 /* don't overrun this mapping */
1831 if (addr + len >= vma->vm_end)
1832 len = vma->vm_end - addr;
1833
1834 /* only read or write mappings where it is permitted */
David Howellsd00c7b992006-09-27 01:50:19 -07001835 if (write && vma->vm_flags & VM_MAYWRITE)
Jie Zhang79597222010-01-06 17:23:28 +00001836 copy_to_user_page(vma, NULL, addr,
1837 (void *) addr, buf, len);
David Howellsd00c7b992006-09-27 01:50:19 -07001838 else if (!write && vma->vm_flags & VM_MAYREAD)
Jie Zhang79597222010-01-06 17:23:28 +00001839 copy_from_user_page(vma, NULL, addr,
1840 buf, (void *) addr, len);
David Howells0ec76a12006-09-27 01:50:15 -07001841 else
1842 len = 0;
1843 } else {
1844 len = 0;
1845 }
1846
1847 up_read(&mm->mmap_sem);
Mike Frysingerf55f1992011-03-29 14:05:12 +01001848
1849 return len;
1850}
1851
1852/**
1853 * @access_remote_vm - access another process' address space
1854 * @mm: the mm_struct of the target address space
1855 * @addr: start address to access
1856 * @buf: source or destination buffer
1857 * @len: number of bytes to transfer
1858 * @write: whether the access is a write
1859 *
1860 * The caller must hold a reference on @mm.
1861 */
1862int access_remote_vm(struct mm_struct *mm, unsigned long addr,
1863 void *buf, int len, int write)
1864{
1865 return __access_remote_vm(NULL, mm, addr, buf, len, write);
1866}
1867
1868/*
1869 * Access another process' address space.
1870 * - source/target buffer must be kernel space
1871 */
1872int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
1873{
1874 struct mm_struct *mm;
1875
1876 if (addr + len < addr)
1877 return 0;
1878
1879 mm = get_task_mm(tsk);
1880 if (!mm)
1881 return 0;
1882
1883 len = __access_remote_vm(tsk, mm, addr, buf, len, write);
1884
David Howells0ec76a12006-09-27 01:50:15 -07001885 mmput(mm);
1886 return len;
1887}
David Howells7e660872010-01-15 17:01:39 -08001888
1889/**
1890 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
1891 * @inode: The inode to check
1892 * @size: The current filesize of the inode
1893 * @newsize: The proposed filesize of the inode
1894 *
1895 * Check the shared mappings on an inode on behalf of a shrinking truncate to
1896 * make sure that that any outstanding VMAs aren't broken and then shrink the
1897 * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
1898 * automatically grant mappings that are too large.
1899 */
1900int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
1901 size_t newsize)
1902{
1903 struct vm_area_struct *vma;
David Howells7e660872010-01-15 17:01:39 -08001904 struct vm_region *region;
1905 pgoff_t low, high;
1906 size_t r_size, r_top;
1907
1908 low = newsize >> PAGE_SHIFT;
1909 high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1910
1911 down_write(&nommu_region_sem);
Davidlohr Bueso1acf2e02014-12-12 16:54:39 -08001912 i_mmap_lock_read(inode->i_mapping);
David Howells7e660872010-01-15 17:01:39 -08001913
1914 /* search for VMAs that fall within the dead zone */
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -07001915 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) {
David Howells7e660872010-01-15 17:01:39 -08001916 /* found one - only interested if it's shared out of the page
1917 * cache */
1918 if (vma->vm_flags & VM_SHARED) {
Davidlohr Bueso1acf2e02014-12-12 16:54:39 -08001919 i_mmap_unlock_read(inode->i_mapping);
David Howells7e660872010-01-15 17:01:39 -08001920 up_write(&nommu_region_sem);
1921 return -ETXTBSY; /* not quite true, but near enough */
1922 }
1923 }
1924
1925 /* reduce any regions that overlap the dead zone - if in existence,
1926 * these will be pointed to by VMAs that don't overlap the dead zone
1927 *
1928 * we don't check for any regions that start beyond the EOF as there
1929 * shouldn't be any
1930 */
Davidlohr Bueso1acf2e02014-12-12 16:54:39 -08001931 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, 0, ULONG_MAX) {
David Howells7e660872010-01-15 17:01:39 -08001932 if (!(vma->vm_flags & VM_SHARED))
1933 continue;
1934
1935 region = vma->vm_region;
1936 r_size = region->vm_top - region->vm_start;
1937 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
1938
1939 if (r_top > newsize) {
1940 region->vm_top -= r_top - newsize;
1941 if (region->vm_end > region->vm_top)
1942 region->vm_end = region->vm_top;
1943 }
1944 }
1945
Davidlohr Bueso1acf2e02014-12-12 16:54:39 -08001946 i_mmap_unlock_read(inode->i_mapping);
David Howells7e660872010-01-15 17:01:39 -08001947 up_write(&nommu_region_sem);
1948 return 0;
1949}
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07001950
1951/*
1952 * Initialise sysctl_user_reserve_kbytes.
1953 *
1954 * This is intended to prevent a user from starting a single memory hogging
1955 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
1956 * mode.
1957 *
1958 * The default value is min(3% of free memory, 128MB)
1959 * 128MB is enough to recover with sshd/login, bash, and top/kill.
1960 */
1961static int __meminit init_user_reserve(void)
1962{
1963 unsigned long free_kbytes;
1964
1965 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
1966
1967 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
1968 return 0;
1969}
Paul Gortmakera4bc6fc2015-05-01 20:08:20 -04001970subsys_initcall(init_user_reserve);
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07001971
1972/*
1973 * Initialise sysctl_admin_reserve_kbytes.
1974 *
1975 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
1976 * to log in and kill a memory hogging process.
1977 *
1978 * Systems with more than 256MB will reserve 8MB, enough to recover
1979 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
1980 * only reserve 3% of free pages by default.
1981 */
1982static int __meminit init_admin_reserve(void)
1983{
1984 unsigned long free_kbytes;
1985
1986 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
1987
1988 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
1989 return 0;
1990}
Paul Gortmakera4bc6fc2015-05-01 20:08:20 -04001991subsys_initcall(init_admin_reserve);
Dave Hansencde70142016-02-12 13:01:55 -08001992
1993long get_user_pages8(struct task_struct *tsk, struct mm_struct *mm,
1994 unsigned long start, unsigned long nr_pages,
1995 int write, int force, struct page **pages,
1996 struct vm_area_struct **vmas)
1997{
1998 return get_user_pages6(start, nr_pages, write, force, pages, vmas);
1999}
2000EXPORT_SYMBOL(get_user_pages8);
2001
2002long get_user_pages_locked8(struct task_struct *tsk, struct mm_struct *mm,
2003 unsigned long start, unsigned long nr_pages,
2004 int write, int force, struct page **pages,
2005 int *locked)
2006{
2007 return get_user_pages_locked6(start, nr_pages, write,
2008 force, pages, locked);
2009}
2010EXPORT_SYMBOL(get_user_pages_locked8);
2011
2012long get_user_pages_unlocked7(struct task_struct *tsk, struct mm_struct *mm,
2013 unsigned long start, unsigned long nr_pages,
2014 int write, int force, struct page **pages)
2015{
2016 return get_user_pages_unlocked5(start, nr_pages, write, force, pages);
2017}
2018EXPORT_SYMBOL(get_user_pages_unlocked7);
2019