blob: 92be862c859bdce2b149e86760522ccb6d230524 [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>
Clark Williamscf4aebc22013-02-07 09:46:59 -060036#include <linux/sched/sysctl.h>
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -070037#include <linux/printk.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038
39#include <asm/uaccess.h>
40#include <asm/tlb.h>
41#include <asm/tlbflush.h>
Bernd Schmidteb8cdec2009-09-21 17:03:57 -070042#include <asm/mmu_context.h>
David Howells8feae132009-01-08 12:04:47 +000043#include "internal.h"
44
Linus Torvalds1da177e2005-04-16 15:20:36 -070045void *high_memory;
Arnd Bergmann944b6872015-02-05 12:25:12 -080046EXPORT_SYMBOL(high_memory);
Linus Torvalds1da177e2005-04-16 15:20:36 -070047struct page *mem_map;
48unsigned long max_mapnr;
gchen gchen5b8bf302015-03-12 16:26:05 -070049EXPORT_SYMBOL(max_mapnr);
Hugh Dickins4266c972009-09-23 17:05:53 +010050unsigned long highest_memmap_pfn;
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -070051struct percpu_counter vm_committed_as;
Linus Torvalds1da177e2005-04-16 15:20:36 -070052int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
53int sysctl_overcommit_ratio = 50; /* default is 50% */
Jerome Marchand49f0ce52014-01-21 15:49:14 -080054unsigned long sysctl_overcommit_kbytes __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070055int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
David Howellsfc4d5c22009-05-06 16:03:05 -070056int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -070057unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -070058unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
Linus Torvalds1da177e2005-04-16 15:20:36 -070059int heap_stack_gap = 0;
60
David Howells33e5d7692009-04-02 16:56:32 -070061atomic_long_t mmap_pages_allocated;
David Howells8feae132009-01-08 12:04:47 +000062
K. Y. Srinivasan997071b2012-11-15 14:34:42 -080063/*
64 * The global memory commitment made in the system can be a metric
65 * that can be used to drive ballooning decisions when Linux is hosted
66 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
67 * balancing memory across competing virtual machines that are hosted.
68 * Several metrics drive this policy engine including the guest reported
69 * memory commitment.
70 */
71unsigned long vm_memory_committed(void)
72{
73 return percpu_counter_read_positive(&vm_committed_as);
74}
75
76EXPORT_SYMBOL_GPL(vm_memory_committed);
77
Linus Torvalds1da177e2005-04-16 15:20:36 -070078EXPORT_SYMBOL(mem_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -070079
David Howells8feae132009-01-08 12:04:47 +000080/* list of mapped, potentially shareable regions */
81static struct kmem_cache *vm_region_jar;
82struct rb_root nommu_region_tree = RB_ROOT;
83DECLARE_RWSEM(nommu_region_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -070084
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +040085const struct vm_operations_struct generic_file_vm_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -070086};
87
88/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070089 * Return the total memory allocated for this pointer, not
90 * just what the caller asked for.
91 *
92 * Doesn't have to be accurate, i.e. may have races.
93 */
94unsigned int kobjsize(const void *objp)
95{
96 struct page *page;
97
Michael Hennerich4016a132008-04-28 02:13:38 -070098 /*
99 * If the object we have should not have ksize performed on it,
100 * return size of 0
101 */
Paul Mundt5a1603b2008-06-12 16:29:55 +0900102 if (!objp || !virt_addr_valid(objp))
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700103 return 0;
104
105 page = virt_to_head_page(objp);
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700106
107 /*
108 * If the allocator sets PageSlab, we know the pointer came from
109 * kmalloc().
110 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 if (PageSlab(page))
112 return ksize(objp);
113
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700114 /*
Paul Mundtab2e83e2009-01-08 12:04:48 +0000115 * If it's not a compound page, see if we have a matching VMA
116 * region. This test is intentionally done in reverse order,
117 * so if there's no VMA, we still fall through and hand back
118 * PAGE_SIZE for 0-order pages.
119 */
120 if (!PageCompound(page)) {
121 struct vm_area_struct *vma;
122
123 vma = find_vma(current->mm, (unsigned long)objp);
124 if (vma)
125 return vma->vm_end - vma->vm_start;
126 }
127
128 /*
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700129 * The ksize() function is only guaranteed to work for pointers
Paul Mundt5a1603b2008-06-12 16:29:55 +0900130 * returned by kmalloc(). So handle arbitrary pointers here.
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700131 */
Paul Mundt5a1603b2008-06-12 16:29:55 +0900132 return PAGE_SIZE << compound_order(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133}
134
Michel Lespinasse28a35712013-02-22 16:35:55 -0800135long __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
136 unsigned long start, unsigned long nr_pages,
137 unsigned int foll_flags, struct page **pages,
138 struct vm_area_struct **vmas, int *nonblocking)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139{
Sonic Zhang910e46d2006-09-27 01:50:17 -0700140 struct vm_area_struct *vma;
David Howells7b4d5b82006-09-27 01:50:18 -0700141 unsigned long vm_flags;
142 int i;
143
144 /* calculate required read or write permissions.
Hugh Dickins58fa8792009-09-21 17:03:31 -0700145 * If FOLL_FORCE is set, we only require the "MAY" flags.
David Howells7b4d5b82006-09-27 01:50:18 -0700146 */
Hugh Dickins58fa8792009-09-21 17:03:31 -0700147 vm_flags = (foll_flags & FOLL_WRITE) ?
148 (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
149 vm_flags &= (foll_flags & FOLL_FORCE) ?
150 (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151
Peter Zijlstra9d737772009-06-25 11:58:55 +0200152 for (i = 0; i < nr_pages; i++) {
David Howells7561e8c2010-03-25 16:48:38 +0000153 vma = find_vma(mm, start);
David Howells7b4d5b82006-09-27 01:50:18 -0700154 if (!vma)
155 goto finish_or_fault;
156
157 /* protect what we can, including chardevs */
Hugh Dickins1c3aff12009-09-21 17:03:24 -0700158 if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
159 !(vm_flags & vma->vm_flags))
David Howells7b4d5b82006-09-27 01:50:18 -0700160 goto finish_or_fault;
Sonic Zhang910e46d2006-09-27 01:50:17 -0700161
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 if (pages) {
163 pages[i] = virt_to_page(start);
164 if (pages[i])
165 page_cache_get(pages[i]);
166 }
167 if (vmas)
Sonic Zhang910e46d2006-09-27 01:50:17 -0700168 vmas[i] = vma;
David Howellse1ee65d2010-03-25 16:48:44 +0000169 start = (start + PAGE_SIZE) & PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 }
David Howells7b4d5b82006-09-27 01:50:18 -0700171
172 return i;
173
174finish_or_fault:
175 return i ? : -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176}
Nick Pigginb291f002008-10-18 20:26:44 -0700177
Nick Pigginb291f002008-10-18 20:26:44 -0700178/*
179 * get a list of pages in an address range belonging to the specified process
180 * and indicate the VMA that covers each page
181 * - this is potentially dodgy as we may end incrementing the page count of a
182 * slab page or a secondary page from a compound page
183 * - don't permit access to VMAs that don't support it, such as I/O mappings
184 */
Michel Lespinasse28a35712013-02-22 16:35:55 -0800185long get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
186 unsigned long start, unsigned long nr_pages,
187 int write, int force, struct page **pages,
188 struct vm_area_struct **vmas)
Nick Pigginb291f002008-10-18 20:26:44 -0700189{
190 int flags = 0;
191
192 if (write)
Hugh Dickins58fa8792009-09-21 17:03:31 -0700193 flags |= FOLL_WRITE;
Nick Pigginb291f002008-10-18 20:26:44 -0700194 if (force)
Hugh Dickins58fa8792009-09-21 17:03:31 -0700195 flags |= FOLL_FORCE;
Nick Pigginb291f002008-10-18 20:26:44 -0700196
Michel Lespinasse53a77062011-01-13 15:46:14 -0800197 return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas,
198 NULL);
Nick Pigginb291f002008-10-18 20:26:44 -0700199}
Greg Ungerer66aa2b42005-09-12 11:18:10 +1000200EXPORT_SYMBOL(get_user_pages);
201
Andrea Arcangelif0818f42015-02-11 15:27:17 -0800202long get_user_pages_locked(struct task_struct *tsk, struct mm_struct *mm,
203 unsigned long start, unsigned long nr_pages,
204 int write, int force, struct page **pages,
205 int *locked)
206{
207 return get_user_pages(tsk, mm, start, nr_pages, write, force,
208 pages, NULL);
209}
210EXPORT_SYMBOL(get_user_pages_locked);
211
Andrea Arcangeli0fd71a52015-02-11 15:27:20 -0800212long __get_user_pages_unlocked(struct task_struct *tsk, struct mm_struct *mm,
213 unsigned long start, unsigned long nr_pages,
214 int write, int force, struct page **pages,
215 unsigned int gup_flags)
Andrea Arcangelif0818f42015-02-11 15:27:17 -0800216{
217 long ret;
218 down_read(&mm->mmap_sem);
219 ret = get_user_pages(tsk, mm, start, nr_pages, write, force,
220 pages, NULL);
221 up_read(&mm->mmap_sem);
222 return ret;
223}
Andrea Arcangeli0fd71a52015-02-11 15:27:20 -0800224EXPORT_SYMBOL(__get_user_pages_unlocked);
225
226long get_user_pages_unlocked(struct task_struct *tsk, struct mm_struct *mm,
227 unsigned long start, unsigned long nr_pages,
228 int write, int force, struct page **pages)
229{
230 return __get_user_pages_unlocked(tsk, mm, start, nr_pages, write,
231 force, pages, 0);
232}
Andrea Arcangelif0818f42015-02-11 15:27:17 -0800233EXPORT_SYMBOL(get_user_pages_unlocked);
234
Paul Mundtdfc2f912009-06-26 04:31:57 +0900235/**
236 * follow_pfn - look up PFN at a user virtual address
237 * @vma: memory mapping
238 * @address: user virtual address
239 * @pfn: location to store found PFN
240 *
241 * Only IO mappings and raw PFN mappings are allowed.
242 *
243 * Returns zero and the pfn at @pfn on success, -ve otherwise.
244 */
245int follow_pfn(struct vm_area_struct *vma, unsigned long address,
246 unsigned long *pfn)
247{
248 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
249 return -EINVAL;
250
251 *pfn = address >> PAGE_SHIFT;
252 return 0;
253}
254EXPORT_SYMBOL(follow_pfn);
255
Joonsoo Kimf1c40692013-04-29 15:07:37 -0700256LIST_HEAD(vmap_area_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800258void vfree(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259{
260 kfree(addr);
261}
Paul Mundtb5073172007-07-21 04:37:25 -0700262EXPORT_SYMBOL(vfree);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263
Al Virodd0fc662005-10-07 07:46:04 +0100264void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265{
266 /*
Robert P. J. Day85186092007-10-19 23:11:38 +0200267 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
268 * returns only a logical address.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 */
Nick Piggin84097512006-03-22 00:08:34 -0800270 return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271}
Paul Mundtb5073172007-07-21 04:37:25 -0700272EXPORT_SYMBOL(__vmalloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273
Paul Mundtf905bc42008-02-04 22:29:59 -0800274void *vmalloc_user(unsigned long size)
275{
276 void *ret;
277
278 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
279 PAGE_KERNEL);
280 if (ret) {
281 struct vm_area_struct *vma;
282
283 down_write(&current->mm->mmap_sem);
284 vma = find_vma(current->mm, (unsigned long)ret);
285 if (vma)
286 vma->vm_flags |= VM_USERMAP;
287 up_write(&current->mm->mmap_sem);
288 }
289
290 return ret;
291}
292EXPORT_SYMBOL(vmalloc_user);
293
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800294struct page *vmalloc_to_page(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295{
296 return virt_to_page(addr);
297}
Paul Mundtb5073172007-07-21 04:37:25 -0700298EXPORT_SYMBOL(vmalloc_to_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800300unsigned long vmalloc_to_pfn(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301{
302 return page_to_pfn(virt_to_page(addr));
303}
Paul Mundtb5073172007-07-21 04:37:25 -0700304EXPORT_SYMBOL(vmalloc_to_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305
306long vread(char *buf, char *addr, unsigned long count)
307{
Chen Gang9bde9162013-07-03 15:02:36 -0700308 /* Don't allow overflow */
309 if ((unsigned long) buf + count < count)
310 count = -(unsigned long) buf;
311
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 memcpy(buf, addr, count);
313 return count;
314}
315
316long vwrite(char *buf, char *addr, unsigned long count)
317{
318 /* Don't allow overflow */
319 if ((unsigned long) addr + count < count)
320 count = -(unsigned long) addr;
321
322 memcpy(addr, buf, count);
Choi Gi-yongac714902014-04-07 15:37:36 -0700323 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324}
325
326/*
Masahiro Yamadae1c05062015-07-07 10:14:59 +0900327 * vmalloc - allocate virtually contiguous memory
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 *
329 * @size: allocation size
330 *
331 * Allocate enough pages to cover @size from the page level
Masahiro Yamadae1c05062015-07-07 10:14:59 +0900332 * allocator and map them into contiguous kernel virtual space.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +0200334 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 * use __vmalloc() instead.
336 */
337void *vmalloc(unsigned long size)
338{
339 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
340}
Andrew Mortonf6138882006-02-28 16:59:18 -0800341EXPORT_SYMBOL(vmalloc);
342
Dave Younge1ca7782010-10-26 14:22:06 -0700343/*
Masahiro Yamadae1c05062015-07-07 10:14:59 +0900344 * vzalloc - allocate virtually contiguous memory with zero fill
Dave Younge1ca7782010-10-26 14:22:06 -0700345 *
346 * @size: allocation size
347 *
348 * Allocate enough pages to cover @size from the page level
Masahiro Yamadae1c05062015-07-07 10:14:59 +0900349 * allocator and map them into contiguous kernel virtual space.
Dave Younge1ca7782010-10-26 14:22:06 -0700350 * The memory allocated is set to zero.
351 *
352 * For tight control over page level allocator and protection flags
353 * use __vmalloc() instead.
354 */
355void *vzalloc(unsigned long size)
356{
357 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
358 PAGE_KERNEL);
359}
360EXPORT_SYMBOL(vzalloc);
361
362/**
363 * vmalloc_node - allocate memory on a specific node
364 * @size: allocation size
365 * @node: numa node
366 *
367 * Allocate enough pages to cover @size from the page level
368 * allocator and map them into contiguous kernel virtual space.
369 *
370 * For tight control over page level allocator and protection flags
371 * use __vmalloc() instead.
372 */
Andrew Mortonf6138882006-02-28 16:59:18 -0800373void *vmalloc_node(unsigned long size, int node)
374{
375 return vmalloc(size);
376}
Paul Mundt9a14f652010-12-24 11:50:34 +0900377EXPORT_SYMBOL(vmalloc_node);
Dave Younge1ca7782010-10-26 14:22:06 -0700378
379/**
380 * vzalloc_node - allocate memory on a specific node with zero fill
381 * @size: allocation size
382 * @node: numa node
383 *
384 * Allocate enough pages to cover @size from the page level
385 * allocator and map them into contiguous kernel virtual space.
386 * The memory allocated is set to zero.
387 *
388 * For tight control over page level allocator and protection flags
389 * use __vmalloc() instead.
390 */
391void *vzalloc_node(unsigned long size, int node)
392{
393 return vzalloc(size);
394}
395EXPORT_SYMBOL(vzalloc_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396
Paul Mundt1af446e2008-08-04 16:01:47 +0900397#ifndef PAGE_KERNEL_EXEC
398# define PAGE_KERNEL_EXEC PAGE_KERNEL
399#endif
400
401/**
402 * vmalloc_exec - allocate virtually contiguous, executable memory
403 * @size: allocation size
404 *
405 * Kernel-internal function to allocate enough pages to cover @size
406 * the page level allocator and map them into contiguous and
407 * executable kernel virtual space.
408 *
409 * For tight control over page level allocator and protection flags
410 * use __vmalloc() instead.
411 */
412
413void *vmalloc_exec(unsigned long size)
414{
415 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
416}
417
Paul Mundtb5073172007-07-21 04:37:25 -0700418/**
419 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 * @size: allocation size
421 *
422 * Allocate enough 32bit PA addressable pages to cover @size from the
Masahiro Yamadae1c05062015-07-07 10:14:59 +0900423 * page level allocator and map them into contiguous kernel virtual space.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 */
425void *vmalloc_32(unsigned long size)
426{
427 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
428}
Paul Mundtb5073172007-07-21 04:37:25 -0700429EXPORT_SYMBOL(vmalloc_32);
430
431/**
432 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
433 * @size: allocation size
434 *
435 * The resulting memory area is 32bit addressable and zeroed so it can be
436 * mapped to userspace without leaking data.
Paul Mundtf905bc42008-02-04 22:29:59 -0800437 *
438 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
439 * remap_vmalloc_range() are permissible.
Paul Mundtb5073172007-07-21 04:37:25 -0700440 */
441void *vmalloc_32_user(unsigned long size)
442{
Paul Mundtf905bc42008-02-04 22:29:59 -0800443 /*
444 * We'll have to sort out the ZONE_DMA bits for 64-bit,
445 * but for now this can simply use vmalloc_user() directly.
446 */
447 return vmalloc_user(size);
Paul Mundtb5073172007-07-21 04:37:25 -0700448}
449EXPORT_SYMBOL(vmalloc_32_user);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450
451void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
452{
453 BUG();
454 return NULL;
455}
Paul Mundtb5073172007-07-21 04:37:25 -0700456EXPORT_SYMBOL(vmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800458void vunmap(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459{
460 BUG();
461}
Paul Mundtb5073172007-07-21 04:37:25 -0700462EXPORT_SYMBOL(vunmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463
Paul Mundteb6434d2009-01-21 17:45:47 +0900464void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
465{
466 BUG();
467 return NULL;
468}
469EXPORT_SYMBOL(vm_map_ram);
470
471void vm_unmap_ram(const void *mem, unsigned int count)
472{
473 BUG();
474}
475EXPORT_SYMBOL(vm_unmap_ram);
476
477void vm_unmap_aliases(void)
478{
479}
480EXPORT_SYMBOL_GPL(vm_unmap_aliases);
481
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482/*
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -0700483 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
484 * have one.
485 */
Gideon Israel Dsouza3b321232014-04-07 15:37:26 -0700486void __weak vmalloc_sync_all(void)
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -0700487{
488}
489
Paul Mundt29c185e2010-12-24 12:08:30 +0900490/**
491 * alloc_vm_area - allocate a range of kernel address space
492 * @size: size of the area
493 *
494 * Returns: NULL on failure, vm_struct on success
495 *
496 * This function reserves a range of kernel address space, and
497 * allocates pagetables to map that range. No actual mappings
498 * are created. If the kernel address space is not shared
499 * between processes, it syncs the pagetable across all
500 * processes.
501 */
David Vrabelcd129092011-09-29 16:53:32 +0100502struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
Paul Mundt29c185e2010-12-24 12:08:30 +0900503{
504 BUG();
505 return NULL;
506}
507EXPORT_SYMBOL_GPL(alloc_vm_area);
508
509void free_vm_area(struct vm_struct *area)
510{
511 BUG();
512}
513EXPORT_SYMBOL_GPL(free_vm_area);
514
Paul Mundtb5073172007-07-21 04:37:25 -0700515int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
516 struct page *page)
517{
518 return -EINVAL;
519}
520EXPORT_SYMBOL(vm_insert_page);
521
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -0700522/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 * sys_brk() for the most part doesn't need the global kernel
524 * lock, except when an application is doing something nasty
525 * like trying to un-brk an area that has already been mapped
526 * to a regular file. in this case, the unmapping will need
527 * to invoke file system routines that need the global lock.
528 */
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100529SYSCALL_DEFINE1(brk, unsigned long, brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530{
531 struct mm_struct *mm = current->mm;
532
533 if (brk < mm->start_brk || brk > mm->context.end_brk)
534 return mm->brk;
535
536 if (mm->brk == brk)
537 return mm->brk;
538
539 /*
540 * Always allow shrinking brk
541 */
542 if (brk <= mm->brk) {
543 mm->brk = brk;
544 return brk;
545 }
546
547 /*
548 * Ok, looks good - let it rip.
549 */
Mike Frysingercfe79c02010-01-06 17:23:23 +0000550 flush_icache_range(mm->brk, brk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 return mm->brk = brk;
552}
553
David Howells8feae132009-01-08 12:04:47 +0000554/*
555 * initialise the VMA and region record slabs
556 */
557void __init mmap_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558{
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -0700559 int ret;
560
Tejun Heo908c7f12014-09-08 09:51:29 +0900561 ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -0700562 VM_BUG_ON(ret);
David Howells33e5d7692009-04-02 16:56:32 -0700563 vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC);
David Howells8feae132009-01-08 12:04:47 +0000564}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565
David Howells8feae132009-01-08 12:04:47 +0000566/*
567 * validate the region tree
568 * - the caller must hold the region lock
569 */
570#ifdef CONFIG_DEBUG_NOMMU_REGIONS
571static noinline void validate_nommu_regions(void)
572{
573 struct vm_region *region, *last;
574 struct rb_node *p, *lastp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575
David Howells8feae132009-01-08 12:04:47 +0000576 lastp = rb_first(&nommu_region_tree);
577 if (!lastp)
578 return;
579
580 last = rb_entry(lastp, struct vm_region, vm_rb);
Geliang Tangc9427bc2015-11-05 18:48:38 -0800581 BUG_ON(last->vm_end <= last->vm_start);
582 BUG_ON(last->vm_top < last->vm_end);
David Howells8feae132009-01-08 12:04:47 +0000583
584 while ((p = rb_next(lastp))) {
585 region = rb_entry(p, struct vm_region, vm_rb);
586 last = rb_entry(lastp, struct vm_region, vm_rb);
587
Geliang Tangc9427bc2015-11-05 18:48:38 -0800588 BUG_ON(region->vm_end <= region->vm_start);
589 BUG_ON(region->vm_top < region->vm_end);
590 BUG_ON(region->vm_start < last->vm_top);
David Howells8feae132009-01-08 12:04:47 +0000591
592 lastp = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 }
594}
David Howells8feae132009-01-08 12:04:47 +0000595#else
David Howells33e5d7692009-04-02 16:56:32 -0700596static void validate_nommu_regions(void)
597{
598}
David Howells8feae132009-01-08 12:04:47 +0000599#endif
600
601/*
602 * add a region into the global tree
603 */
604static void add_nommu_region(struct vm_region *region)
605{
606 struct vm_region *pregion;
607 struct rb_node **p, *parent;
608
609 validate_nommu_regions();
610
David Howells8feae132009-01-08 12:04:47 +0000611 parent = NULL;
612 p = &nommu_region_tree.rb_node;
613 while (*p) {
614 parent = *p;
615 pregion = rb_entry(parent, struct vm_region, vm_rb);
616 if (region->vm_start < pregion->vm_start)
617 p = &(*p)->rb_left;
618 else if (region->vm_start > pregion->vm_start)
619 p = &(*p)->rb_right;
620 else if (pregion == region)
621 return;
622 else
623 BUG();
624 }
625
626 rb_link_node(&region->vm_rb, parent, p);
627 rb_insert_color(&region->vm_rb, &nommu_region_tree);
628
629 validate_nommu_regions();
630}
631
632/*
633 * delete a region from the global tree
634 */
635static void delete_nommu_region(struct vm_region *region)
636{
637 BUG_ON(!nommu_region_tree.rb_node);
638
639 validate_nommu_regions();
640 rb_erase(&region->vm_rb, &nommu_region_tree);
641 validate_nommu_regions();
642}
643
644/*
645 * free a contiguous series of pages
646 */
647static void free_page_series(unsigned long from, unsigned long to)
648{
649 for (; from < to; from += PAGE_SIZE) {
650 struct page *page = virt_to_page(from);
651
David Howells33e5d7692009-04-02 16:56:32 -0700652 atomic_long_dec(&mmap_pages_allocated);
David Howells8feae132009-01-08 12:04:47 +0000653 put_page(page);
654 }
655}
656
657/*
658 * release a reference to a region
David Howells33e5d7692009-04-02 16:56:32 -0700659 * - the caller must hold the region semaphore for writing, which this releases
Paul Mundtdd8632a2009-01-08 12:04:47 +0000660 * - the region may not have been added to the tree yet, in which case vm_top
David Howells8feae132009-01-08 12:04:47 +0000661 * will equal vm_start
662 */
663static void __put_nommu_region(struct vm_region *region)
664 __releases(nommu_region_sem)
665{
David Howells8feae132009-01-08 12:04:47 +0000666 BUG_ON(!nommu_region_tree.rb_node);
667
David Howells1e2ae592010-01-15 17:01:33 -0800668 if (--region->vm_usage == 0) {
Paul Mundtdd8632a2009-01-08 12:04:47 +0000669 if (region->vm_top > region->vm_start)
David Howells8feae132009-01-08 12:04:47 +0000670 delete_nommu_region(region);
671 up_write(&nommu_region_sem);
672
673 if (region->vm_file)
674 fput(region->vm_file);
675
676 /* IO memory and memory shared directly out of the pagecache
677 * from ramfs/tmpfs mustn't be released here */
Leon Romanovsky22cc8772015-06-24 16:57:47 -0700678 if (region->vm_flags & VM_MAPPED_COPY)
Paul Mundtdd8632a2009-01-08 12:04:47 +0000679 free_page_series(region->vm_start, region->vm_top);
David Howells8feae132009-01-08 12:04:47 +0000680 kmem_cache_free(vm_region_jar, region);
681 } else {
682 up_write(&nommu_region_sem);
683 }
684}
685
686/*
687 * release a reference to a region
688 */
689static void put_nommu_region(struct vm_region *region)
690{
691 down_write(&nommu_region_sem);
692 __put_nommu_region(region);
693}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694
David Howells30340972006-09-27 01:50:20 -0700695/*
Bernd Schmidteb8cdec2009-09-21 17:03:57 -0700696 * update protection on a vma
697 */
698static void protect_vma(struct vm_area_struct *vma, unsigned long flags)
699{
700#ifdef CONFIG_MPU
701 struct mm_struct *mm = vma->vm_mm;
702 long start = vma->vm_start & PAGE_MASK;
703 while (start < vma->vm_end) {
704 protect_page(mm, start, flags);
705 start += PAGE_SIZE;
706 }
707 update_protections(mm);
708#endif
709}
710
711/*
David Howells30340972006-09-27 01:50:20 -0700712 * add a VMA into a process's mm_struct in the appropriate place in the list
David Howells8feae132009-01-08 12:04:47 +0000713 * and tree and add to the address space's page tree also if not an anonymous
714 * page
David Howells30340972006-09-27 01:50:20 -0700715 * - should be called with mm->mmap_sem held writelocked
716 */
David Howells8feae132009-01-08 12:04:47 +0000717static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
David Howells30340972006-09-27 01:50:20 -0700718{
Namhyung Kim6038def2011-05-24 17:11:22 -0700719 struct vm_area_struct *pvma, *prev;
David Howells8feae132009-01-08 12:04:47 +0000720 struct address_space *mapping;
Namhyung Kim6038def2011-05-24 17:11:22 -0700721 struct rb_node **p, *parent, *rb_prev;
David Howells30340972006-09-27 01:50:20 -0700722
David Howells8feae132009-01-08 12:04:47 +0000723 BUG_ON(!vma->vm_region);
724
725 mm->map_count++;
726 vma->vm_mm = mm;
727
Bernd Schmidteb8cdec2009-09-21 17:03:57 -0700728 protect_vma(vma, vma->vm_flags);
729
David Howells8feae132009-01-08 12:04:47 +0000730 /* add the VMA to the mapping */
731 if (vma->vm_file) {
732 mapping = vma->vm_file->f_mapping;
733
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800734 i_mmap_lock_write(mapping);
David Howells8feae132009-01-08 12:04:47 +0000735 flush_dcache_mmap_lock(mapping);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700736 vma_interval_tree_insert(vma, &mapping->i_mmap);
David Howells8feae132009-01-08 12:04:47 +0000737 flush_dcache_mmap_unlock(mapping);
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800738 i_mmap_unlock_write(mapping);
David Howells8feae132009-01-08 12:04:47 +0000739 }
740
741 /* add the VMA to the tree */
Namhyung Kim6038def2011-05-24 17:11:22 -0700742 parent = rb_prev = NULL;
David Howells8feae132009-01-08 12:04:47 +0000743 p = &mm->mm_rb.rb_node;
744 while (*p) {
745 parent = *p;
746 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
747
748 /* sort by: start addr, end addr, VMA struct addr in that order
749 * (the latter is necessary as we may get identical VMAs) */
750 if (vma->vm_start < pvma->vm_start)
751 p = &(*p)->rb_left;
Namhyung Kim6038def2011-05-24 17:11:22 -0700752 else if (vma->vm_start > pvma->vm_start) {
753 rb_prev = parent;
David Howells8feae132009-01-08 12:04:47 +0000754 p = &(*p)->rb_right;
Namhyung Kim6038def2011-05-24 17:11:22 -0700755 } else if (vma->vm_end < pvma->vm_end)
David Howells8feae132009-01-08 12:04:47 +0000756 p = &(*p)->rb_left;
Namhyung Kim6038def2011-05-24 17:11:22 -0700757 else if (vma->vm_end > pvma->vm_end) {
758 rb_prev = parent;
David Howells8feae132009-01-08 12:04:47 +0000759 p = &(*p)->rb_right;
Namhyung Kim6038def2011-05-24 17:11:22 -0700760 } else if (vma < pvma)
David Howells8feae132009-01-08 12:04:47 +0000761 p = &(*p)->rb_left;
Namhyung Kim6038def2011-05-24 17:11:22 -0700762 else if (vma > pvma) {
763 rb_prev = parent;
David Howells8feae132009-01-08 12:04:47 +0000764 p = &(*p)->rb_right;
Namhyung Kim6038def2011-05-24 17:11:22 -0700765 } else
David Howells8feae132009-01-08 12:04:47 +0000766 BUG();
767 }
768
769 rb_link_node(&vma->vm_rb, parent, p);
770 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
771
772 /* add VMA to the VMA list also */
Namhyung Kim6038def2011-05-24 17:11:22 -0700773 prev = NULL;
774 if (rb_prev)
775 prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
David Howells30340972006-09-27 01:50:20 -0700776
Namhyung Kim6038def2011-05-24 17:11:22 -0700777 __vma_link_list(mm, vma, prev, parent);
David Howells8feae132009-01-08 12:04:47 +0000778}
779
780/*
781 * delete a VMA from its owning mm_struct and address space
782 */
783static void delete_vma_from_mm(struct vm_area_struct *vma)
784{
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700785 int i;
David Howells8feae132009-01-08 12:04:47 +0000786 struct address_space *mapping;
787 struct mm_struct *mm = vma->vm_mm;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700788 struct task_struct *curr = current;
David Howells8feae132009-01-08 12:04:47 +0000789
Bernd Schmidteb8cdec2009-09-21 17:03:57 -0700790 protect_vma(vma, 0);
791
David Howells8feae132009-01-08 12:04:47 +0000792 mm->map_count--;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700793 for (i = 0; i < VMACACHE_SIZE; i++) {
794 /* if the vma is cached, invalidate the entire cache */
795 if (curr->vmacache[i] == vma) {
Steven Miaoe020d5b2014-06-23 13:22:02 -0700796 vmacache_invalidate(mm);
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700797 break;
798 }
799 }
David Howells8feae132009-01-08 12:04:47 +0000800
801 /* remove the VMA from the mapping */
802 if (vma->vm_file) {
803 mapping = vma->vm_file->f_mapping;
804
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800805 i_mmap_lock_write(mapping);
David Howells8feae132009-01-08 12:04:47 +0000806 flush_dcache_mmap_lock(mapping);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700807 vma_interval_tree_remove(vma, &mapping->i_mmap);
David Howells8feae132009-01-08 12:04:47 +0000808 flush_dcache_mmap_unlock(mapping);
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800809 i_mmap_unlock_write(mapping);
David Howells8feae132009-01-08 12:04:47 +0000810 }
811
812 /* remove from the MM's tree and list */
813 rb_erase(&vma->vm_rb, &mm->mm_rb);
Namhyung Kimb951bf22011-05-24 17:11:23 -0700814
815 if (vma->vm_prev)
816 vma->vm_prev->vm_next = vma->vm_next;
817 else
818 mm->mmap = vma->vm_next;
819
820 if (vma->vm_next)
821 vma->vm_next->vm_prev = vma->vm_prev;
David Howells8feae132009-01-08 12:04:47 +0000822}
823
824/*
825 * destroy a VMA record
826 */
827static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
828{
David Howells8feae132009-01-08 12:04:47 +0000829 if (vma->vm_ops && vma->vm_ops->close)
830 vma->vm_ops->close(vma);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -0700831 if (vma->vm_file)
David Howells8feae132009-01-08 12:04:47 +0000832 fput(vma->vm_file);
David Howells8feae132009-01-08 12:04:47 +0000833 put_nommu_region(vma->vm_region);
834 kmem_cache_free(vm_area_cachep, vma);
David Howells30340972006-09-27 01:50:20 -0700835}
836
837/*
838 * look up the first VMA in which addr resides, NULL if none
839 * - should be called with mm->mmap_sem at least held readlocked
840 */
841struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
842{
David Howells8feae132009-01-08 12:04:47 +0000843 struct vm_area_struct *vma;
David Howells30340972006-09-27 01:50:20 -0700844
David Howells8feae132009-01-08 12:04:47 +0000845 /* check the cache first */
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700846 vma = vmacache_find(mm, addr);
847 if (likely(vma))
David Howells8feae132009-01-08 12:04:47 +0000848 return vma;
849
Namhyung Kime922c4c2011-05-24 17:11:24 -0700850 /* trawl the list (there may be multiple mappings in which addr
David Howells8feae132009-01-08 12:04:47 +0000851 * resides) */
Namhyung Kime922c4c2011-05-24 17:11:24 -0700852 for (vma = mm->mmap; vma; vma = vma->vm_next) {
David Howells8feae132009-01-08 12:04:47 +0000853 if (vma->vm_start > addr)
854 return NULL;
855 if (vma->vm_end > addr) {
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700856 vmacache_update(addr, vma);
David Howells8feae132009-01-08 12:04:47 +0000857 return vma;
858 }
David Howells30340972006-09-27 01:50:20 -0700859 }
860
David Howells30340972006-09-27 01:50:20 -0700861 return NULL;
862}
863EXPORT_SYMBOL(find_vma);
864
865/*
David Howells930e6522006-09-27 01:50:22 -0700866 * find a VMA
867 * - we don't extend stack VMAs under NOMMU conditions
868 */
869struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
870{
David Howells7561e8c2010-03-25 16:48:38 +0000871 return find_vma(mm, addr);
David Howells930e6522006-09-27 01:50:22 -0700872}
873
David Howells8feae132009-01-08 12:04:47 +0000874/*
875 * expand a stack to a given address
876 * - not supported under NOMMU conditions
877 */
Greg Ungerer57c8f632007-07-15 23:38:28 -0700878int expand_stack(struct vm_area_struct *vma, unsigned long address)
879{
880 return -ENOMEM;
881}
882
David Howells930e6522006-09-27 01:50:22 -0700883/*
David Howells6fa5f802006-09-27 01:50:21 -0700884 * look up the first VMA exactly that exactly matches addr
885 * - should be called with mm->mmap_sem at least held readlocked
886 */
David Howells8feae132009-01-08 12:04:47 +0000887static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
888 unsigned long addr,
889 unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890{
891 struct vm_area_struct *vma;
David Howells8feae132009-01-08 12:04:47 +0000892 unsigned long end = addr + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893
David Howells8feae132009-01-08 12:04:47 +0000894 /* check the cache first */
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700895 vma = vmacache_find_exact(mm, addr, end);
896 if (vma)
David Howells8feae132009-01-08 12:04:47 +0000897 return vma;
898
Namhyung Kime922c4c2011-05-24 17:11:24 -0700899 /* trawl the list (there may be multiple mappings in which addr
David Howells8feae132009-01-08 12:04:47 +0000900 * resides) */
Namhyung Kime922c4c2011-05-24 17:11:24 -0700901 for (vma = mm->mmap; vma; vma = vma->vm_next) {
David Howells8feae132009-01-08 12:04:47 +0000902 if (vma->vm_start < addr)
903 continue;
904 if (vma->vm_start > addr)
905 return NULL;
906 if (vma->vm_end == end) {
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700907 vmacache_update(addr, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 return vma;
David Howells8feae132009-01-08 12:04:47 +0000909 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 }
911
912 return NULL;
913}
914
David Howells30340972006-09-27 01:50:20 -0700915/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 * determine whether a mapping should be permitted and, if so, what sort of
917 * mapping we're capable of supporting
918 */
919static int validate_mmap_request(struct file *file,
920 unsigned long addr,
921 unsigned long len,
922 unsigned long prot,
923 unsigned long flags,
924 unsigned long pgoff,
925 unsigned long *_capabilities)
926{
David Howells8feae132009-01-08 12:04:47 +0000927 unsigned long capabilities, rlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928 int ret;
929
930 /* do the simple checks first */
Leon Romanovsky22cc8772015-06-24 16:57:47 -0700931 if (flags & MAP_FIXED)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933
934 if ((flags & MAP_TYPE) != MAP_PRIVATE &&
935 (flags & MAP_TYPE) != MAP_SHARED)
936 return -EINVAL;
937
Mike Frysingerf81cff02006-12-06 12:02:59 +1000938 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 return -EINVAL;
940
Mike Frysingerf81cff02006-12-06 12:02:59 +1000941 /* Careful about overflows.. */
David Howells8feae132009-01-08 12:04:47 +0000942 rlen = PAGE_ALIGN(len);
943 if (!rlen || rlen > TASK_SIZE)
Mike Frysingerf81cff02006-12-06 12:02:59 +1000944 return -ENOMEM;
945
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 /* offset overflow? */
David Howells8feae132009-01-08 12:04:47 +0000947 if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
Mike Frysingerf81cff02006-12-06 12:02:59 +1000948 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949
950 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951 /* files must support mmap */
Al Viro72c2d532013-09-22 16:27:52 -0400952 if (!file->f_op->mmap)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 return -ENODEV;
954
955 /* work out if what we've got could possibly be shared
956 * - we support chardevs that provide their own "memory"
957 * - we support files/blockdevs that are memory backed
958 */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100959 if (file->f_op->mmap_capabilities) {
960 capabilities = file->f_op->mmap_capabilities(file);
961 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 /* no explicit capabilities set, so assume some
963 * defaults */
Al Viro496ad9a2013-01-23 17:07:38 -0500964 switch (file_inode(file)->i_mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 case S_IFREG:
966 case S_IFBLK:
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100967 capabilities = NOMMU_MAP_COPY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 break;
969
970 case S_IFCHR:
971 capabilities =
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100972 NOMMU_MAP_DIRECT |
973 NOMMU_MAP_READ |
974 NOMMU_MAP_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 break;
976
977 default:
978 return -EINVAL;
979 }
980 }
981
982 /* eliminate any capabilities that we can't support on this
983 * device */
984 if (!file->f_op->get_unmapped_area)
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100985 capabilities &= ~NOMMU_MAP_DIRECT;
Al Viro6e242a12015-03-31 12:35:13 -0400986 if (!(file->f_mode & FMODE_CAN_READ))
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100987 capabilities &= ~NOMMU_MAP_COPY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988
Graff Yang28d7a6a2009-08-18 14:11:17 -0700989 /* The file shall have been opened with read permission. */
990 if (!(file->f_mode & FMODE_READ))
991 return -EACCES;
992
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 if (flags & MAP_SHARED) {
994 /* do checks for writing, appending and locking */
995 if ((prot & PROT_WRITE) &&
996 !(file->f_mode & FMODE_WRITE))
997 return -EACCES;
998
Al Viro496ad9a2013-01-23 17:07:38 -0500999 if (IS_APPEND(file_inode(file)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 (file->f_mode & FMODE_WRITE))
1001 return -EACCES;
1002
Jeff Laytond7a06982014-03-10 09:54:15 -04001003 if (locks_verify_locked(file))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 return -EAGAIN;
1005
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001006 if (!(capabilities & NOMMU_MAP_DIRECT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 return -ENODEV;
1008
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 /* we mustn't privatise shared mappings */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001010 capabilities &= ~NOMMU_MAP_COPY;
Choi Gi-yongac714902014-04-07 15:37:36 -07001011 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 /* we're going to read the file into private memory we
1013 * allocate */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001014 if (!(capabilities & NOMMU_MAP_COPY))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 return -ENODEV;
1016
1017 /* we don't permit a private writable mapping to be
1018 * shared with the backing device */
1019 if (prot & PROT_WRITE)
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001020 capabilities &= ~NOMMU_MAP_DIRECT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 }
1022
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001023 if (capabilities & NOMMU_MAP_DIRECT) {
1024 if (((prot & PROT_READ) && !(capabilities & NOMMU_MAP_READ)) ||
1025 ((prot & PROT_WRITE) && !(capabilities & NOMMU_MAP_WRITE)) ||
1026 ((prot & PROT_EXEC) && !(capabilities & NOMMU_MAP_EXEC))
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001027 ) {
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001028 capabilities &= ~NOMMU_MAP_DIRECT;
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001029 if (flags & MAP_SHARED) {
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001030 pr_warn("MAP_SHARED not completely supported on !MMU\n");
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001031 return -EINVAL;
1032 }
1033 }
1034 }
1035
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 /* handle executable mappings and implied executable
1037 * mappings */
Eric W. Biederman90f85722015-06-29 14:42:03 -05001038 if (path_noexec(&file->f_path)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 if (prot & PROT_EXEC)
1040 return -EPERM;
Choi Gi-yongac714902014-04-07 15:37:36 -07001041 } else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 /* handle implication of PROT_EXEC by PROT_READ */
1043 if (current->personality & READ_IMPLIES_EXEC) {
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001044 if (capabilities & NOMMU_MAP_EXEC)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 prot |= PROT_EXEC;
1046 }
Choi Gi-yongac714902014-04-07 15:37:36 -07001047 } else if ((prot & PROT_READ) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 (prot & PROT_EXEC) &&
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001049 !(capabilities & NOMMU_MAP_EXEC)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 ) {
1051 /* backing file is not executable, try to copy */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001052 capabilities &= ~NOMMU_MAP_DIRECT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 }
Choi Gi-yongac714902014-04-07 15:37:36 -07001054 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 /* anonymous mappings are always memory backed and can be
1056 * privately mapped
1057 */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001058 capabilities = NOMMU_MAP_COPY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059
1060 /* handle PROT_EXEC implication by PROT_READ */
1061 if ((prot & PROT_READ) &&
1062 (current->personality & READ_IMPLIES_EXEC))
1063 prot |= PROT_EXEC;
1064 }
1065
1066 /* allow the security API to have its say */
Al Viroe5467852012-05-30 13:30:51 -04001067 ret = security_mmap_addr(addr);
1068 if (ret < 0)
1069 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070
1071 /* looks okay */
1072 *_capabilities = capabilities;
1073 return 0;
1074}
1075
1076/*
1077 * we've determined that we can make the mapping, now translate what we
1078 * now know into VMA flags
1079 */
1080static unsigned long determine_vm_flags(struct file *file,
1081 unsigned long prot,
1082 unsigned long flags,
1083 unsigned long capabilities)
1084{
1085 unsigned long vm_flags;
1086
1087 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 /* vm_flags |= mm->def_flags; */
1089
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001090 if (!(capabilities & NOMMU_MAP_DIRECT)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 /* attempt to share read-only copies of mapped file chunks */
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001092 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 if (file && !(prot & PROT_WRITE))
1094 vm_flags |= VM_MAYSHARE;
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001095 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 /* overlay a shareable mapping on the backing device or inode
1097 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1098 * romfs/cramfs */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001099 vm_flags |= VM_MAYSHARE | (capabilities & NOMMU_VMFLAGS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 if (flags & MAP_SHARED)
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001101 vm_flags |= VM_SHARED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 }
1103
1104 /* refuse to let anyone share private mappings with this process if
1105 * it's being traced - otherwise breakpoints set in it may interfere
1106 * with another untraced process
1107 */
Tejun Heoa288eec2011-06-17 16:50:37 +02001108 if ((flags & MAP_PRIVATE) && current->ptrace)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109 vm_flags &= ~VM_MAYSHARE;
1110
1111 return vm_flags;
1112}
1113
1114/*
David Howells8feae132009-01-08 12:04:47 +00001115 * set up a shared mapping on a file (the driver or filesystem provides and
1116 * pins the storage)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 */
David Howells8feae132009-01-08 12:04:47 +00001118static int do_mmap_shared_file(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119{
1120 int ret;
1121
1122 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001123 if (ret == 0) {
1124 vma->vm_region->vm_top = vma->vm_region->vm_end;
David Howells645d83c2009-09-24 15:13:10 +01001125 return 0;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001126 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 if (ret != -ENOSYS)
1128 return ret;
1129
David Howells3fa30462010-03-23 13:35:21 -07001130 /* getting -ENOSYS indicates that direct mmap isn't possible (as
1131 * opposed to tried but failed) so we can only give a suitable error as
1132 * it's not possible to make a private copy if MAP_SHARED was given */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 return -ENODEV;
1134}
1135
1136/*
1137 * set up a private mapping or an anonymous shared mapping
1138 */
David Howells8feae132009-01-08 12:04:47 +00001139static int do_mmap_private(struct vm_area_struct *vma,
1140 struct vm_region *region,
David Howells645d83c2009-09-24 15:13:10 +01001141 unsigned long len,
1142 unsigned long capabilities)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143{
Joonsoo Kimdbc83582014-12-12 16:55:55 -08001144 unsigned long total, point;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 void *base;
David Howells8feae132009-01-08 12:04:47 +00001146 int ret, order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147
1148 /* invoke the file's mapping function so that it can keep track of
1149 * shared mappings on devices or memory
1150 * - VM_MAYSHARE will be set if it may attempt to share
1151 */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001152 if (capabilities & NOMMU_MAP_DIRECT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001154 if (ret == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155 /* shouldn't return success if we're not sharing */
Paul Mundtdd8632a2009-01-08 12:04:47 +00001156 BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1157 vma->vm_region->vm_top = vma->vm_region->vm_end;
David Howells645d83c2009-09-24 15:13:10 +01001158 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 }
Paul Mundtdd8632a2009-01-08 12:04:47 +00001160 if (ret != -ENOSYS)
1161 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162
1163 /* getting an ENOSYS error indicates that direct mmap isn't
1164 * possible (as opposed to tried but failed) so we'll try to
1165 * make a private copy of the data and map that instead */
1166 }
1167
David Howells8feae132009-01-08 12:04:47 +00001168
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 /* allocate some memory to hold the mapping
1170 * - note that this may not return a page-aligned address if the object
1171 * we're allocating is smaller than a page
1172 */
Bob Liuf67d9b12011-05-24 17:12:56 -07001173 order = get_order(len);
David Howells8feae132009-01-08 12:04:47 +00001174 total = 1 << order;
Bob Liuf67d9b12011-05-24 17:12:56 -07001175 point = len >> PAGE_SHIFT;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001176
Joonsoo Kimdbc83582014-12-12 16:55:55 -08001177 /* we don't want to allocate a power-of-2 sized page set */
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001178 if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages)
Joonsoo Kimdbc83582014-12-12 16:55:55 -08001179 total = point;
David Howells8feae132009-01-08 12:04:47 +00001180
Joonsoo Kimda616532015-02-27 15:51:43 -08001181 base = alloc_pages_exact(total << PAGE_SHIFT, GFP_KERNEL);
Joonsoo Kimdbc83582014-12-12 16:55:55 -08001182 if (!base)
1183 goto enomem;
David Howells8feae132009-01-08 12:04:47 +00001184
Joonsoo Kimdbc83582014-12-12 16:55:55 -08001185 atomic_long_add(total, &mmap_pages_allocated);
1186
David Howells8feae132009-01-08 12:04:47 +00001187 region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1188 region->vm_start = (unsigned long) base;
Bob Liuf67d9b12011-05-24 17:12:56 -07001189 region->vm_end = region->vm_start + len;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001190 region->vm_top = region->vm_start + (total << PAGE_SHIFT);
David Howells8feae132009-01-08 12:04:47 +00001191
1192 vma->vm_start = region->vm_start;
1193 vma->vm_end = region->vm_start + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194
1195 if (vma->vm_file) {
1196 /* read the contents of a file into the copy */
1197 mm_segment_t old_fs;
1198 loff_t fpos;
1199
1200 fpos = vma->vm_pgoff;
1201 fpos <<= PAGE_SHIFT;
1202
1203 old_fs = get_fs();
1204 set_fs(KERNEL_DS);
Al Viro6e242a12015-03-31 12:35:13 -04001205 ret = __vfs_read(vma->vm_file, base, len, &fpos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 set_fs(old_fs);
1207
1208 if (ret < 0)
1209 goto error_free;
1210
1211 /* clear the last little bit */
Bob Liuf67d9b12011-05-24 17:12:56 -07001212 if (ret < len)
1213 memset(base + ret, 0, len - ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 }
1216
1217 return 0;
1218
1219error_free:
Namhyung Kim7223bb42011-05-24 17:11:26 -07001220 free_page_series(region->vm_start, region->vm_top);
David Howells8feae132009-01-08 12:04:47 +00001221 region->vm_start = vma->vm_start = 0;
1222 region->vm_end = vma->vm_end = 0;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001223 region->vm_top = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 return ret;
1225
1226enomem:
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -07001227 pr_err("Allocation of length %lu from process %d (%s) failed\n",
Greg Ungerer05ae6fa2009-01-13 17:30:22 +10001228 len, current->pid, current->comm);
David Rientjes7bf02ea2011-05-24 17:11:16 -07001229 show_free_areas(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 return -ENOMEM;
1231}
1232
1233/*
1234 * handle mapping creation for uClinux
1235 */
Oleg Nesterov1fcfd8d2015-09-09 15:39:29 -07001236unsigned long do_mmap(struct file *file,
1237 unsigned long addr,
1238 unsigned long len,
1239 unsigned long prot,
1240 unsigned long flags,
1241 vm_flags_t vm_flags,
1242 unsigned long pgoff,
1243 unsigned long *populate)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244{
David Howells8feae132009-01-08 12:04:47 +00001245 struct vm_area_struct *vma;
1246 struct vm_region *region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247 struct rb_node *rb;
Oleg Nesterov1fcfd8d2015-09-09 15:39:29 -07001248 unsigned long capabilities, result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249 int ret;
1250
Michel Lespinasse41badc12013-02-22 16:32:47 -08001251 *populate = 0;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001252
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 /* decide whether we should attempt the mapping, and if so what sort of
1254 * mapping */
1255 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1256 &capabilities);
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001257 if (ret < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258 return ret;
1259
David Howells06aab5a2009-09-24 12:33:48 +01001260 /* we ignore the address hint */
1261 addr = 0;
Bob Liuf67d9b12011-05-24 17:12:56 -07001262 len = PAGE_ALIGN(len);
David Howells06aab5a2009-09-24 12:33:48 +01001263
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 /* we've determined that we can make the mapping, now translate what we
1265 * now know into VMA flags */
Oleg Nesterov1fcfd8d2015-09-09 15:39:29 -07001266 vm_flags |= determine_vm_flags(file, prot, flags, capabilities);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267
David Howells8feae132009-01-08 12:04:47 +00001268 /* we're going to need to record the mapping */
1269 region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1270 if (!region)
1271 goto error_getting_region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272
David Howells8feae132009-01-08 12:04:47 +00001273 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1274 if (!vma)
1275 goto error_getting_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276
David Howells1e2ae592010-01-15 17:01:33 -08001277 region->vm_usage = 1;
David Howells8feae132009-01-08 12:04:47 +00001278 region->vm_flags = vm_flags;
1279 region->vm_pgoff = pgoff;
1280
Rik van Riel5beb4932010-03-05 13:42:07 -08001281 INIT_LIST_HEAD(&vma->anon_vma_chain);
David Howells8feae132009-01-08 12:04:47 +00001282 vma->vm_flags = vm_flags;
1283 vma->vm_pgoff = pgoff;
1284
1285 if (file) {
Al Virocb0942b2012-08-27 14:48:26 -04001286 region->vm_file = get_file(file);
1287 vma->vm_file = get_file(file);
David Howells8feae132009-01-08 12:04:47 +00001288 }
1289
1290 down_write(&nommu_region_sem);
1291
1292 /* if we want to share, we need to check for regions created by other
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 * mmap() calls that overlap with our proposed mapping
David Howells8feae132009-01-08 12:04:47 +00001294 * - we can only share with a superset match on most regular files
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295 * - shared mappings on character devices and memory backed files are
1296 * permitted to overlap inexactly as far as we are concerned for in
1297 * these cases, sharing is handled in the driver or filesystem rather
1298 * than here
1299 */
1300 if (vm_flags & VM_MAYSHARE) {
David Howells8feae132009-01-08 12:04:47 +00001301 struct vm_region *pregion;
1302 unsigned long pglen, rpglen, pgend, rpgend, start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303
David Howells8feae132009-01-08 12:04:47 +00001304 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1305 pgend = pgoff + pglen;
David Howells165b2392007-03-22 00:11:24 -08001306
David Howells8feae132009-01-08 12:04:47 +00001307 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1308 pregion = rb_entry(rb, struct vm_region, vm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309
David Howells8feae132009-01-08 12:04:47 +00001310 if (!(pregion->vm_flags & VM_MAYSHARE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311 continue;
1312
1313 /* search for overlapping mappings on the same file */
Al Viro496ad9a2013-01-23 17:07:38 -05001314 if (file_inode(pregion->vm_file) !=
1315 file_inode(file))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 continue;
1317
David Howells8feae132009-01-08 12:04:47 +00001318 if (pregion->vm_pgoff >= pgend)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319 continue;
1320
David Howells8feae132009-01-08 12:04:47 +00001321 rpglen = pregion->vm_end - pregion->vm_start;
1322 rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1323 rpgend = pregion->vm_pgoff + rpglen;
1324 if (pgoff >= rpgend)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 continue;
1326
David Howells8feae132009-01-08 12:04:47 +00001327 /* handle inexactly overlapping matches between
1328 * mappings */
1329 if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1330 !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1331 /* new mapping is not a subset of the region */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001332 if (!(capabilities & NOMMU_MAP_DIRECT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333 goto sharing_violation;
1334 continue;
1335 }
1336
David Howells8feae132009-01-08 12:04:47 +00001337 /* we've found a region we can share */
David Howells1e2ae592010-01-15 17:01:33 -08001338 pregion->vm_usage++;
David Howells8feae132009-01-08 12:04:47 +00001339 vma->vm_region = pregion;
1340 start = pregion->vm_start;
1341 start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1342 vma->vm_start = start;
1343 vma->vm_end = start + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001345 if (pregion->vm_flags & VM_MAPPED_COPY)
David Howells8feae132009-01-08 12:04:47 +00001346 vma->vm_flags |= VM_MAPPED_COPY;
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001347 else {
David Howells8feae132009-01-08 12:04:47 +00001348 ret = do_mmap_shared_file(vma);
1349 if (ret < 0) {
1350 vma->vm_region = NULL;
1351 vma->vm_start = 0;
1352 vma->vm_end = 0;
David Howells1e2ae592010-01-15 17:01:33 -08001353 pregion->vm_usage--;
David Howells8feae132009-01-08 12:04:47 +00001354 pregion = NULL;
1355 goto error_just_free;
1356 }
1357 }
1358 fput(region->vm_file);
1359 kmem_cache_free(vm_region_jar, region);
1360 region = pregion;
1361 result = start;
1362 goto share;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 }
1364
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 /* obtain the address at which to make a shared mapping
1366 * - this is the hook for quasi-memory character devices to
1367 * tell us the location of a shared mapping
1368 */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001369 if (capabilities & NOMMU_MAP_DIRECT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 addr = file->f_op->get_unmapped_area(file, addr, len,
1371 pgoff, flags);
Namhyung Kimbb005a52011-05-24 17:11:27 -07001372 if (IS_ERR_VALUE(addr)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 ret = addr;
Namhyung Kimbb005a52011-05-24 17:11:27 -07001374 if (ret != -ENOSYS)
David Howells8feae132009-01-08 12:04:47 +00001375 goto error_just_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376
1377 /* the driver refused to tell us where to site
1378 * the mapping so we'll have to attempt to copy
1379 * it */
Namhyung Kimbb005a52011-05-24 17:11:27 -07001380 ret = -ENODEV;
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001381 if (!(capabilities & NOMMU_MAP_COPY))
David Howells8feae132009-01-08 12:04:47 +00001382 goto error_just_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001384 capabilities &= ~NOMMU_MAP_DIRECT;
David Howells8feae132009-01-08 12:04:47 +00001385 } else {
1386 vma->vm_start = region->vm_start = addr;
1387 vma->vm_end = region->vm_end = addr + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 }
1389 }
1390 }
1391
David Howells8feae132009-01-08 12:04:47 +00001392 vma->vm_region = region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393
David Howells645d83c2009-09-24 15:13:10 +01001394 /* set up the mapping
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001395 * - the region is filled in if NOMMU_MAP_DIRECT is still set
David Howells645d83c2009-09-24 15:13:10 +01001396 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 if (file && vma->vm_flags & VM_SHARED)
David Howells8feae132009-01-08 12:04:47 +00001398 ret = do_mmap_shared_file(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 else
David Howells645d83c2009-09-24 15:13:10 +01001400 ret = do_mmap_private(vma, region, len, capabilities);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 if (ret < 0)
David Howells645d83c2009-09-24 15:13:10 +01001402 goto error_just_free;
1403 add_nommu_region(region);
David Howells8feae132009-01-08 12:04:47 +00001404
Jie Zhangea637632009-12-14 18:00:02 -08001405 /* clear anonymous mappings that don't ask for uninitialized data */
1406 if (!vma->vm_file && !(flags & MAP_UNINITIALIZED))
1407 memset((void *)region->vm_start, 0,
1408 region->vm_end - region->vm_start);
1409
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 /* okay... we have a mapping; now we have to register it */
David Howells8feae132009-01-08 12:04:47 +00001411 result = vma->vm_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 current->mm->total_vm += len >> PAGE_SHIFT;
1414
David Howells8feae132009-01-08 12:04:47 +00001415share:
1416 add_vma_to_mm(current->mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417
Mike Frysingercfe79c02010-01-06 17:23:23 +00001418 /* we flush the region from the icache only when the first executable
1419 * mapping of it is made */
1420 if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1421 flush_icache_range(region->vm_start, region->vm_end);
1422 region->vm_icache_flushed = true;
1423 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424
Mike Frysingercfe79c02010-01-06 17:23:23 +00001425 up_write(&nommu_region_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426
David Howells8feae132009-01-08 12:04:47 +00001427 return result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428
David Howells8feae132009-01-08 12:04:47 +00001429error_just_free:
1430 up_write(&nommu_region_sem);
1431error:
David Howells89a86402009-10-30 13:13:26 +00001432 if (region->vm_file)
1433 fput(region->vm_file);
David Howells8feae132009-01-08 12:04:47 +00001434 kmem_cache_free(vm_region_jar, region);
David Howells89a86402009-10-30 13:13:26 +00001435 if (vma->vm_file)
1436 fput(vma->vm_file);
David Howells8feae132009-01-08 12:04:47 +00001437 kmem_cache_free(vm_area_cachep, vma);
David Howells8feae132009-01-08 12:04:47 +00001438 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439
David Howells8feae132009-01-08 12:04:47 +00001440sharing_violation:
1441 up_write(&nommu_region_sem);
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001442 pr_warn("Attempt to share mismatched mappings\n");
David Howells8feae132009-01-08 12:04:47 +00001443 ret = -EINVAL;
1444 goto error;
1445
1446error_getting_vma:
1447 kmem_cache_free(vm_region_jar, region);
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001448 pr_warn("Allocation of vma for %lu byte allocation from process %d failed\n",
1449 len, current->pid);
David Rientjes7bf02ea2011-05-24 17:11:16 -07001450 show_free_areas(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451 return -ENOMEM;
1452
David Howells8feae132009-01-08 12:04:47 +00001453error_getting_region:
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001454 pr_warn("Allocation of vm region for %lu byte allocation from process %d failed\n",
1455 len, current->pid);
David Rientjes7bf02ea2011-05-24 17:11:16 -07001456 show_free_areas(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 return -ENOMEM;
1458}
Linus Torvalds6be5ceb2012-04-20 17:13:58 -07001459
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001460SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1461 unsigned long, prot, unsigned long, flags,
1462 unsigned long, fd, unsigned long, pgoff)
1463{
1464 struct file *file = NULL;
1465 unsigned long retval = -EBADF;
1466
Al Viro120a7952010-10-30 02:54:44 -04001467 audit_mmap_fd(fd, flags);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001468 if (!(flags & MAP_ANONYMOUS)) {
1469 file = fget(fd);
1470 if (!file)
1471 goto out;
1472 }
1473
1474 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1475
Greg Ungererad1ed292012-06-04 14:29:59 +10001476 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001477
1478 if (file)
1479 fput(file);
1480out:
1481 return retval;
1482}
1483
Christoph Hellwiga4679372010-03-10 15:21:15 -08001484#ifdef __ARCH_WANT_SYS_OLD_MMAP
1485struct mmap_arg_struct {
1486 unsigned long addr;
1487 unsigned long len;
1488 unsigned long prot;
1489 unsigned long flags;
1490 unsigned long fd;
1491 unsigned long offset;
1492};
1493
1494SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1495{
1496 struct mmap_arg_struct a;
1497
1498 if (copy_from_user(&a, arg, sizeof(a)))
1499 return -EFAULT;
Alexander Kuleshov1824cb72015-11-05 18:46:35 -08001500 if (offset_in_page(a.offset))
Christoph Hellwiga4679372010-03-10 15:21:15 -08001501 return -EINVAL;
1502
1503 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1504 a.offset >> PAGE_SHIFT);
1505}
1506#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1507
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508/*
David Howells8feae132009-01-08 12:04:47 +00001509 * split a vma into two pieces at address 'addr', a new vma is allocated either
1510 * for the first part or the tail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 */
David Howells8feae132009-01-08 12:04:47 +00001512int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1513 unsigned long addr, int new_below)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514{
David Howells8feae132009-01-08 12:04:47 +00001515 struct vm_area_struct *new;
1516 struct vm_region *region;
1517 unsigned long npages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518
David Howells779c1022010-01-15 17:01:34 -08001519 /* we're only permitted to split anonymous regions (these should have
1520 * only a single usage on the region) */
1521 if (vma->vm_file)
David Howells8feae132009-01-08 12:04:47 +00001522 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523
David Howells8feae132009-01-08 12:04:47 +00001524 if (mm->map_count >= sysctl_max_map_count)
1525 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526
David Howells8feae132009-01-08 12:04:47 +00001527 region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1528 if (!region)
1529 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530
David Howells8feae132009-01-08 12:04:47 +00001531 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1532 if (!new) {
1533 kmem_cache_free(vm_region_jar, region);
1534 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535 }
David Howells8feae132009-01-08 12:04:47 +00001536
1537 /* most fields are the same, copy all, and then fixup */
1538 *new = *vma;
1539 *region = *vma->vm_region;
1540 new->vm_region = region;
1541
1542 npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1543
1544 if (new_below) {
Paul Mundtdd8632a2009-01-08 12:04:47 +00001545 region->vm_top = region->vm_end = new->vm_end = addr;
David Howells8feae132009-01-08 12:04:47 +00001546 } else {
1547 region->vm_start = new->vm_start = addr;
1548 region->vm_pgoff = new->vm_pgoff += npages;
1549 }
1550
1551 if (new->vm_ops && new->vm_ops->open)
1552 new->vm_ops->open(new);
1553
1554 delete_vma_from_mm(vma);
1555 down_write(&nommu_region_sem);
1556 delete_nommu_region(vma->vm_region);
1557 if (new_below) {
1558 vma->vm_region->vm_start = vma->vm_start = addr;
1559 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1560 } else {
1561 vma->vm_region->vm_end = vma->vm_end = addr;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001562 vma->vm_region->vm_top = addr;
David Howells8feae132009-01-08 12:04:47 +00001563 }
1564 add_nommu_region(vma->vm_region);
1565 add_nommu_region(new->vm_region);
1566 up_write(&nommu_region_sem);
1567 add_vma_to_mm(mm, vma);
1568 add_vma_to_mm(mm, new);
1569 return 0;
1570}
1571
1572/*
1573 * shrink a VMA by removing the specified chunk from either the beginning or
1574 * the end
1575 */
1576static int shrink_vma(struct mm_struct *mm,
1577 struct vm_area_struct *vma,
1578 unsigned long from, unsigned long to)
1579{
1580 struct vm_region *region;
1581
David Howells8feae132009-01-08 12:04:47 +00001582 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1583 * and list */
1584 delete_vma_from_mm(vma);
1585 if (from > vma->vm_start)
1586 vma->vm_end = from;
1587 else
1588 vma->vm_start = to;
1589 add_vma_to_mm(mm, vma);
1590
1591 /* cut the backing region down to size */
1592 region = vma->vm_region;
David Howells1e2ae592010-01-15 17:01:33 -08001593 BUG_ON(region->vm_usage != 1);
David Howells8feae132009-01-08 12:04:47 +00001594
1595 down_write(&nommu_region_sem);
1596 delete_nommu_region(region);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001597 if (from > region->vm_start) {
1598 to = region->vm_top;
1599 region->vm_top = region->vm_end = from;
1600 } else {
David Howells8feae132009-01-08 12:04:47 +00001601 region->vm_start = to;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001602 }
David Howells8feae132009-01-08 12:04:47 +00001603 add_nommu_region(region);
1604 up_write(&nommu_region_sem);
1605
1606 free_page_series(from, to);
1607 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608}
1609
David Howells30340972006-09-27 01:50:20 -07001610/*
1611 * release a mapping
David Howells8feae132009-01-08 12:04:47 +00001612 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1613 * VMA, though it need not cover the whole VMA
David Howells30340972006-09-27 01:50:20 -07001614 */
David Howells8feae132009-01-08 12:04:47 +00001615int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616{
David Howells8feae132009-01-08 12:04:47 +00001617 struct vm_area_struct *vma;
Bob Liuf67d9b12011-05-24 17:12:56 -07001618 unsigned long end;
David Howells8feae132009-01-08 12:04:47 +00001619 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620
Bob Liuf67d9b12011-05-24 17:12:56 -07001621 len = PAGE_ALIGN(len);
David Howells8feae132009-01-08 12:04:47 +00001622 if (len == 0)
1623 return -EINVAL;
1624
Bob Liuf67d9b12011-05-24 17:12:56 -07001625 end = start + len;
1626
David Howells8feae132009-01-08 12:04:47 +00001627 /* find the first potentially overlapping VMA */
1628 vma = find_vma(mm, start);
1629 if (!vma) {
Choi Gi-yongac714902014-04-07 15:37:36 -07001630 static int limit;
David Howells33e5d7692009-04-02 16:56:32 -07001631 if (limit < 5) {
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001632 pr_warn("munmap of memory not mmapped by process %d (%s): 0x%lx-0x%lx\n",
1633 current->pid, current->comm,
1634 start, start + len - 1);
David Howells33e5d7692009-04-02 16:56:32 -07001635 limit++;
1636 }
David Howells8feae132009-01-08 12:04:47 +00001637 return -EINVAL;
David Howells30340972006-09-27 01:50:20 -07001638 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639
David Howells8feae132009-01-08 12:04:47 +00001640 /* we're allowed to split an anonymous VMA but not a file-backed one */
1641 if (vma->vm_file) {
1642 do {
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001643 if (start > vma->vm_start)
David Howells8feae132009-01-08 12:04:47 +00001644 return -EINVAL;
David Howells8feae132009-01-08 12:04:47 +00001645 if (end == vma->vm_end)
1646 goto erase_whole_vma;
Namhyung Kimd75a3102011-05-24 17:11:25 -07001647 vma = vma->vm_next;
1648 } while (vma);
David Howells8feae132009-01-08 12:04:47 +00001649 return -EINVAL;
1650 } else {
1651 /* the chunk must be a subset of the VMA found */
1652 if (start == vma->vm_start && end == vma->vm_end)
1653 goto erase_whole_vma;
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001654 if (start < vma->vm_start || end > vma->vm_end)
David Howells8feae132009-01-08 12:04:47 +00001655 return -EINVAL;
Alexander Kuleshov1824cb72015-11-05 18:46:35 -08001656 if (offset_in_page(start))
David Howells8feae132009-01-08 12:04:47 +00001657 return -EINVAL;
Alexander Kuleshov1824cb72015-11-05 18:46:35 -08001658 if (end != vma->vm_end && offset_in_page(end))
David Howells8feae132009-01-08 12:04:47 +00001659 return -EINVAL;
David Howells8feae132009-01-08 12:04:47 +00001660 if (start != vma->vm_start && end != vma->vm_end) {
1661 ret = split_vma(mm, vma, start, 1);
Leon Romanovsky22cc8772015-06-24 16:57:47 -07001662 if (ret < 0)
David Howells8feae132009-01-08 12:04:47 +00001663 return ret;
David Howells8feae132009-01-08 12:04:47 +00001664 }
1665 return shrink_vma(mm, vma, start, end);
1666 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667
David Howells8feae132009-01-08 12:04:47 +00001668erase_whole_vma:
1669 delete_vma_from_mm(vma);
1670 delete_vma(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671 return 0;
1672}
Paul Mundtb5073172007-07-21 04:37:25 -07001673EXPORT_SYMBOL(do_munmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674
Al Virobfce2812012-04-20 21:57:04 -04001675int vm_munmap(unsigned long addr, size_t len)
David Howells30340972006-09-27 01:50:20 -07001676{
Al Virobfce2812012-04-20 21:57:04 -04001677 struct mm_struct *mm = current->mm;
David Howells30340972006-09-27 01:50:20 -07001678 int ret;
David Howells30340972006-09-27 01:50:20 -07001679
1680 down_write(&mm->mmap_sem);
1681 ret = do_munmap(mm, addr, len);
1682 up_write(&mm->mmap_sem);
1683 return ret;
1684}
Linus Torvaldsa46ef992012-04-20 16:20:01 -07001685EXPORT_SYMBOL(vm_munmap);
1686
1687SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1688{
Al Virobfce2812012-04-20 21:57:04 -04001689 return vm_munmap(addr, len);
Linus Torvaldsa46ef992012-04-20 16:20:01 -07001690}
David Howells30340972006-09-27 01:50:20 -07001691
1692/*
David Howells8feae132009-01-08 12:04:47 +00001693 * release all the mappings made in a process's VM space
David Howells30340972006-09-27 01:50:20 -07001694 */
David Howells8feae132009-01-08 12:04:47 +00001695void exit_mmap(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696{
David Howells8feae132009-01-08 12:04:47 +00001697 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698
David Howells8feae132009-01-08 12:04:47 +00001699 if (!mm)
1700 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701
David Howells8feae132009-01-08 12:04:47 +00001702 mm->total_vm = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703
David Howells8feae132009-01-08 12:04:47 +00001704 while ((vma = mm->mmap)) {
1705 mm->mmap = vma->vm_next;
1706 delete_vma_from_mm(vma);
1707 delete_vma(mm, vma);
Steven J. Magnani04c34962010-11-24 12:56:54 -08001708 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709 }
1710}
1711
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07001712unsigned long vm_brk(unsigned long addr, unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713{
1714 return -ENOMEM;
1715}
1716
1717/*
David Howells6fa5f802006-09-27 01:50:21 -07001718 * expand (or shrink) an existing mapping, potentially moving it at the same
1719 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720 *
David Howells6fa5f802006-09-27 01:50:21 -07001721 * under NOMMU conditions, we only permit changing a mapping's size, and only
David Howells8feae132009-01-08 12:04:47 +00001722 * as long as it stays within the region allocated by do_mmap_private() and the
1723 * block is not shareable
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724 *
David Howells6fa5f802006-09-27 01:50:21 -07001725 * MREMAP_FIXED is not supported under NOMMU conditions
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726 */
Al Viro4b377ba2013-03-04 10:47:59 -05001727static unsigned long do_mremap(unsigned long addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728 unsigned long old_len, unsigned long new_len,
1729 unsigned long flags, unsigned long new_addr)
1730{
David Howells6fa5f802006-09-27 01:50:21 -07001731 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732
1733 /* insanity checks first */
Bob Liuf67d9b12011-05-24 17:12:56 -07001734 old_len = PAGE_ALIGN(old_len);
1735 new_len = PAGE_ALIGN(new_len);
David Howells8feae132009-01-08 12:04:47 +00001736 if (old_len == 0 || new_len == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737 return (unsigned long) -EINVAL;
1738
Alexander Kuleshov1824cb72015-11-05 18:46:35 -08001739 if (offset_in_page(addr))
David Howells8feae132009-01-08 12:04:47 +00001740 return -EINVAL;
1741
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742 if (flags & MREMAP_FIXED && new_addr != addr)
1743 return (unsigned long) -EINVAL;
1744
David Howells8feae132009-01-08 12:04:47 +00001745 vma = find_vma_exact(current->mm, addr, old_len);
David Howells6fa5f802006-09-27 01:50:21 -07001746 if (!vma)
1747 return (unsigned long) -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748
David Howells6fa5f802006-09-27 01:50:21 -07001749 if (vma->vm_end != vma->vm_start + old_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750 return (unsigned long) -EFAULT;
1751
David Howells6fa5f802006-09-27 01:50:21 -07001752 if (vma->vm_flags & VM_MAYSHARE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753 return (unsigned long) -EPERM;
1754
David Howells8feae132009-01-08 12:04:47 +00001755 if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756 return (unsigned long) -ENOMEM;
1757
1758 /* all checks complete - do it */
David Howells6fa5f802006-09-27 01:50:21 -07001759 vma->vm_end = vma->vm_start + new_len;
David Howells6fa5f802006-09-27 01:50:21 -07001760 return vma->vm_start;
1761}
1762
Heiko Carstens6a6160a2009-01-14 14:14:15 +01001763SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1764 unsigned long, new_len, unsigned long, flags,
1765 unsigned long, new_addr)
David Howells6fa5f802006-09-27 01:50:21 -07001766{
1767 unsigned long ret;
1768
1769 down_write(&current->mm->mmap_sem);
1770 ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1771 up_write(&current->mm->mmap_sem);
1772 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773}
1774
Michel Lespinasse240aade2013-02-22 16:35:56 -08001775struct page *follow_page_mask(struct vm_area_struct *vma,
1776 unsigned long address, unsigned int flags,
1777 unsigned int *page_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778{
Michel Lespinasse240aade2013-02-22 16:35:56 -08001779 *page_mask = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780 return NULL;
1781}
1782
Bob Liu8f3b1322011-07-08 15:39:46 -07001783int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1784 unsigned long pfn, unsigned long size, pgprot_t prot)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785{
Bob Liu8f3b1322011-07-08 15:39:46 -07001786 if (addr != (pfn << PAGE_SHIFT))
1787 return -EINVAL;
1788
Konstantin Khlebnikov314e51b2012-10-08 16:29:02 -07001789 vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
Greg Ungerer66aa2b42005-09-12 11:18:10 +10001790 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791}
Luke Yang22c4af42006-07-14 00:24:09 -07001792EXPORT_SYMBOL(remap_pfn_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793
Linus Torvalds3c0b9de2013-04-27 13:25:38 -07001794int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)
1795{
1796 unsigned long pfn = start >> PAGE_SHIFT;
1797 unsigned long vm_len = vma->vm_end - vma->vm_start;
1798
1799 pfn += vma->vm_pgoff;
1800 return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);
1801}
1802EXPORT_SYMBOL(vm_iomap_memory);
1803
Paul Mundtf905bc42008-02-04 22:29:59 -08001804int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1805 unsigned long pgoff)
1806{
1807 unsigned int size = vma->vm_end - vma->vm_start;
1808
1809 if (!(vma->vm_flags & VM_USERMAP))
1810 return -EINVAL;
1811
1812 vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1813 vma->vm_end = vma->vm_start + size;
1814
1815 return 0;
1816}
1817EXPORT_SYMBOL(remap_vmalloc_range);
1818
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1820 unsigned long len, unsigned long pgoff, unsigned long flags)
1821{
1822 return -ENOMEM;
1823}
1824
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825void unmap_mapping_range(struct address_space *mapping,
1826 loff_t const holebegin, loff_t const holelen,
1827 int even_cows)
1828{
1829}
Luke Yang22c4af42006-07-14 00:24:09 -07001830EXPORT_SYMBOL(unmap_mapping_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831
1832/*
1833 * Check that a process has enough memory to allocate a new virtual
1834 * mapping. 0 means there is enough memory for the allocation to
1835 * succeed and -ENOMEM implies there is not.
1836 *
1837 * We currently support three overcommit policies, which are set via the
1838 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
1839 *
1840 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1841 * Additional code 2002 Jul 20 by Robert Love.
1842 *
1843 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1844 *
1845 * Note this is a helper function intended to be used by LSMs which
1846 * wish to use this logic.
1847 */
Alan Cox34b4e4a2007-08-22 14:01:28 -07001848int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849{
Roman Gushchin8138a672015-02-11 15:28:42 -08001850 long free, allowed, reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851
1852 vm_acct_memory(pages);
1853
1854 /*
1855 * Sometimes we want to use more memory than we have
1856 */
1857 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1858 return 0;
1859
1860 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
Dmitry Finkc15bef32011-07-25 17:12:19 -07001861 free = global_page_state(NR_FREE_PAGES);
1862 free += global_page_state(NR_FILE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863
Dmitry Finkc15bef32011-07-25 17:12:19 -07001864 /*
1865 * shmem pages shouldn't be counted as free in this
1866 * case, they can't be purged, only swapped out, and
1867 * that won't affect the overall amount of available
1868 * memory in the system.
1869 */
1870 free -= global_page_state(NR_SHMEM);
1871
Shaohua Liec8acf22013-02-22 16:34:38 -08001872 free += get_nr_swap_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873
1874 /*
1875 * Any slabs which are created with the
1876 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1877 * which are reclaimable, under pressure. The dentry
1878 * cache and most inode caches should fall into this
1879 */
Christoph Lameter972d1a72006-09-25 23:31:51 -07001880 free += global_page_state(NR_SLAB_RECLAIMABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881
1882 /*
Dmitry Finkc15bef32011-07-25 17:12:19 -07001883 * Leave reserved pages. The pages are not for anonymous pages.
1884 */
1885 if (free <= totalreserve_pages)
1886 goto error;
1887 else
1888 free -= totalreserve_pages;
1889
1890 /*
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07001891 * Reserve some for root
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892 */
1893 if (!cap_sys_admin)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07001894 free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895
1896 if (free > pages)
1897 return 0;
1898
Hideo AOKId5ddc792006-04-10 22:53:01 -07001899 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900 }
1901
Jerome Marchand00619bc2013-11-12 15:08:31 -08001902 allowed = vm_commit_limit();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903 /*
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07001904 * Reserve some 3% for root
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905 */
1906 if (!cap_sys_admin)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07001907 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07001909 /*
1910 * Don't let a single process grow so big a user can't recover
1911 */
1912 if (mm) {
1913 reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
Roman Gushchin8138a672015-02-11 15:28:42 -08001914 allowed -= min_t(long, mm->total_vm / 32, reserve);
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07001915 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07001917 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918 return 0;
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07001919
Hideo AOKId5ddc792006-04-10 22:53:01 -07001920error:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921 vm_unacct_memory(pages);
1922
1923 return -ENOMEM;
1924}
1925
Nick Piggind0217ac2007-07-19 01:47:03 -07001926int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
David Howellsb0e15192006-01-06 00:11:42 -08001927{
1928 BUG();
Nick Piggind0217ac2007-07-19 01:47:03 -07001929 return 0;
David Howellsb0e15192006-01-06 00:11:42 -08001930}
Paul Mundtb5073172007-07-21 04:37:25 -07001931EXPORT_SYMBOL(filemap_fault);
David Howells0ec76a12006-09-27 01:50:15 -07001932
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07001933void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf)
1934{
1935 BUG();
1936}
1937EXPORT_SYMBOL(filemap_map_pages);
1938
Mike Frysingerf55f1992011-03-29 14:05:12 +01001939static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
1940 unsigned long addr, void *buf, int len, int write)
David Howells0ec76a12006-09-27 01:50:15 -07001941{
David Howells0ec76a12006-09-27 01:50:15 -07001942 struct vm_area_struct *vma;
David Howells0ec76a12006-09-27 01:50:15 -07001943
1944 down_read(&mm->mmap_sem);
1945
1946 /* the access must start within one of the target process's mappings */
David Howells0159b142006-09-27 01:50:16 -07001947 vma = find_vma(mm, addr);
1948 if (vma) {
David Howells0ec76a12006-09-27 01:50:15 -07001949 /* don't overrun this mapping */
1950 if (addr + len >= vma->vm_end)
1951 len = vma->vm_end - addr;
1952
1953 /* only read or write mappings where it is permitted */
David Howellsd00c7b992006-09-27 01:50:19 -07001954 if (write && vma->vm_flags & VM_MAYWRITE)
Jie Zhang79597222010-01-06 17:23:28 +00001955 copy_to_user_page(vma, NULL, addr,
1956 (void *) addr, buf, len);
David Howellsd00c7b992006-09-27 01:50:19 -07001957 else if (!write && vma->vm_flags & VM_MAYREAD)
Jie Zhang79597222010-01-06 17:23:28 +00001958 copy_from_user_page(vma, NULL, addr,
1959 buf, (void *) addr, len);
David Howells0ec76a12006-09-27 01:50:15 -07001960 else
1961 len = 0;
1962 } else {
1963 len = 0;
1964 }
1965
1966 up_read(&mm->mmap_sem);
Mike Frysingerf55f1992011-03-29 14:05:12 +01001967
1968 return len;
1969}
1970
1971/**
1972 * @access_remote_vm - access another process' address space
1973 * @mm: the mm_struct of the target address space
1974 * @addr: start address to access
1975 * @buf: source or destination buffer
1976 * @len: number of bytes to transfer
1977 * @write: whether the access is a write
1978 *
1979 * The caller must hold a reference on @mm.
1980 */
1981int access_remote_vm(struct mm_struct *mm, unsigned long addr,
1982 void *buf, int len, int write)
1983{
1984 return __access_remote_vm(NULL, mm, addr, buf, len, write);
1985}
1986
1987/*
1988 * Access another process' address space.
1989 * - source/target buffer must be kernel space
1990 */
1991int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
1992{
1993 struct mm_struct *mm;
1994
1995 if (addr + len < addr)
1996 return 0;
1997
1998 mm = get_task_mm(tsk);
1999 if (!mm)
2000 return 0;
2001
2002 len = __access_remote_vm(tsk, mm, addr, buf, len, write);
2003
David Howells0ec76a12006-09-27 01:50:15 -07002004 mmput(mm);
2005 return len;
2006}
David Howells7e660872010-01-15 17:01:39 -08002007
2008/**
2009 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
2010 * @inode: The inode to check
2011 * @size: The current filesize of the inode
2012 * @newsize: The proposed filesize of the inode
2013 *
2014 * Check the shared mappings on an inode on behalf of a shrinking truncate to
2015 * make sure that that any outstanding VMAs aren't broken and then shrink the
2016 * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
2017 * automatically grant mappings that are too large.
2018 */
2019int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
2020 size_t newsize)
2021{
2022 struct vm_area_struct *vma;
David Howells7e660872010-01-15 17:01:39 -08002023 struct vm_region *region;
2024 pgoff_t low, high;
2025 size_t r_size, r_top;
2026
2027 low = newsize >> PAGE_SHIFT;
2028 high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
2029
2030 down_write(&nommu_region_sem);
Davidlohr Bueso1acf2e02014-12-12 16:54:39 -08002031 i_mmap_lock_read(inode->i_mapping);
David Howells7e660872010-01-15 17:01:39 -08002032
2033 /* search for VMAs that fall within the dead zone */
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -07002034 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) {
David Howells7e660872010-01-15 17:01:39 -08002035 /* found one - only interested if it's shared out of the page
2036 * cache */
2037 if (vma->vm_flags & VM_SHARED) {
Davidlohr Bueso1acf2e02014-12-12 16:54:39 -08002038 i_mmap_unlock_read(inode->i_mapping);
David Howells7e660872010-01-15 17:01:39 -08002039 up_write(&nommu_region_sem);
2040 return -ETXTBSY; /* not quite true, but near enough */
2041 }
2042 }
2043
2044 /* reduce any regions that overlap the dead zone - if in existence,
2045 * these will be pointed to by VMAs that don't overlap the dead zone
2046 *
2047 * we don't check for any regions that start beyond the EOF as there
2048 * shouldn't be any
2049 */
Davidlohr Bueso1acf2e02014-12-12 16:54:39 -08002050 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, 0, ULONG_MAX) {
David Howells7e660872010-01-15 17:01:39 -08002051 if (!(vma->vm_flags & VM_SHARED))
2052 continue;
2053
2054 region = vma->vm_region;
2055 r_size = region->vm_top - region->vm_start;
2056 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
2057
2058 if (r_top > newsize) {
2059 region->vm_top -= r_top - newsize;
2060 if (region->vm_end > region->vm_top)
2061 region->vm_end = region->vm_top;
2062 }
2063 }
2064
Davidlohr Bueso1acf2e02014-12-12 16:54:39 -08002065 i_mmap_unlock_read(inode->i_mapping);
David Howells7e660872010-01-15 17:01:39 -08002066 up_write(&nommu_region_sem);
2067 return 0;
2068}
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07002069
2070/*
2071 * Initialise sysctl_user_reserve_kbytes.
2072 *
2073 * This is intended to prevent a user from starting a single memory hogging
2074 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
2075 * mode.
2076 *
2077 * The default value is min(3% of free memory, 128MB)
2078 * 128MB is enough to recover with sshd/login, bash, and top/kill.
2079 */
2080static int __meminit init_user_reserve(void)
2081{
2082 unsigned long free_kbytes;
2083
2084 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
2085
2086 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
2087 return 0;
2088}
Paul Gortmakera4bc6fc2015-05-01 20:08:20 -04002089subsys_initcall(init_user_reserve);
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07002090
2091/*
2092 * Initialise sysctl_admin_reserve_kbytes.
2093 *
2094 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
2095 * to log in and kill a memory hogging process.
2096 *
2097 * Systems with more than 256MB will reserve 8MB, enough to recover
2098 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
2099 * only reserve 3% of free pages by default.
2100 */
2101static int __meminit init_admin_reserve(void)
2102{
2103 unsigned long free_kbytes;
2104
2105 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
2106
2107 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
2108 return 0;
2109}
Paul Gortmakera4bc6fc2015-05-01 20:08:20 -04002110subsys_initcall(init_admin_reserve);