blob: 7296360fc057e5bbf67b9904501fdbe98a97b1b2 [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
David Howells8feae132009-01-08 12:04:47 +000045#if 0
46#define kenter(FMT, ...) \
47 printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
48#define kleave(FMT, ...) \
49 printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
50#define kdebug(FMT, ...) \
51 printk(KERN_DEBUG "xxx" FMT"yyy\n", ##__VA_ARGS__)
52#else
53#define kenter(FMT, ...) \
54 no_printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
55#define kleave(FMT, ...) \
56 no_printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
57#define kdebug(FMT, ...) \
58 no_printk(KERN_DEBUG FMT"\n", ##__VA_ARGS__)
59#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -070060
61void *high_memory;
Arnd Bergmann944b6872015-02-05 12:25:12 -080062EXPORT_SYMBOL(high_memory);
Linus Torvalds1da177e2005-04-16 15:20:36 -070063struct page *mem_map;
64unsigned long max_mapnr;
Hugh Dickins4266c972009-09-23 17:05:53 +010065unsigned long highest_memmap_pfn;
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -070066struct percpu_counter vm_committed_as;
Linus Torvalds1da177e2005-04-16 15:20:36 -070067int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
68int sysctl_overcommit_ratio = 50; /* default is 50% */
Jerome Marchand49f0ce52014-01-21 15:49:14 -080069unsigned long sysctl_overcommit_kbytes __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070070int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
David Howellsfc4d5c22009-05-06 16:03:05 -070071int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -070072unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -070073unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
Linus Torvalds1da177e2005-04-16 15:20:36 -070074int heap_stack_gap = 0;
75
David Howells33e5d7692009-04-02 16:56:32 -070076atomic_long_t mmap_pages_allocated;
David Howells8feae132009-01-08 12:04:47 +000077
K. Y. Srinivasan997071b2012-11-15 14:34:42 -080078/*
79 * The global memory commitment made in the system can be a metric
80 * that can be used to drive ballooning decisions when Linux is hosted
81 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
82 * balancing memory across competing virtual machines that are hosted.
83 * Several metrics drive this policy engine including the guest reported
84 * memory commitment.
85 */
86unsigned long vm_memory_committed(void)
87{
88 return percpu_counter_read_positive(&vm_committed_as);
89}
90
91EXPORT_SYMBOL_GPL(vm_memory_committed);
92
Linus Torvalds1da177e2005-04-16 15:20:36 -070093EXPORT_SYMBOL(mem_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -070094
David Howells8feae132009-01-08 12:04:47 +000095/* list of mapped, potentially shareable regions */
96static struct kmem_cache *vm_region_jar;
97struct rb_root nommu_region_tree = RB_ROOT;
98DECLARE_RWSEM(nommu_region_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -070099
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +0400100const struct vm_operations_struct generic_file_vm_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101};
102
103/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104 * Return the total memory allocated for this pointer, not
105 * just what the caller asked for.
106 *
107 * Doesn't have to be accurate, i.e. may have races.
108 */
109unsigned int kobjsize(const void *objp)
110{
111 struct page *page;
112
Michael Hennerich4016a132008-04-28 02:13:38 -0700113 /*
114 * If the object we have should not have ksize performed on it,
115 * return size of 0
116 */
Paul Mundt5a1603b2008-06-12 16:29:55 +0900117 if (!objp || !virt_addr_valid(objp))
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700118 return 0;
119
120 page = virt_to_head_page(objp);
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700121
122 /*
123 * If the allocator sets PageSlab, we know the pointer came from
124 * kmalloc().
125 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 if (PageSlab(page))
127 return ksize(objp);
128
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700129 /*
Paul Mundtab2e83e2009-01-08 12:04:48 +0000130 * If it's not a compound page, see if we have a matching VMA
131 * region. This test is intentionally done in reverse order,
132 * so if there's no VMA, we still fall through and hand back
133 * PAGE_SIZE for 0-order pages.
134 */
135 if (!PageCompound(page)) {
136 struct vm_area_struct *vma;
137
138 vma = find_vma(current->mm, (unsigned long)objp);
139 if (vma)
140 return vma->vm_end - vma->vm_start;
141 }
142
143 /*
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700144 * The ksize() function is only guaranteed to work for pointers
Paul Mundt5a1603b2008-06-12 16:29:55 +0900145 * returned by kmalloc(). So handle arbitrary pointers here.
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700146 */
Paul Mundt5a1603b2008-06-12 16:29:55 +0900147 return PAGE_SIZE << compound_order(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148}
149
Michel Lespinasse28a35712013-02-22 16:35:55 -0800150long __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
151 unsigned long start, unsigned long nr_pages,
152 unsigned int foll_flags, struct page **pages,
153 struct vm_area_struct **vmas, int *nonblocking)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154{
Sonic Zhang910e46d2006-09-27 01:50:17 -0700155 struct vm_area_struct *vma;
David Howells7b4d5b82006-09-27 01:50:18 -0700156 unsigned long vm_flags;
157 int i;
158
159 /* calculate required read or write permissions.
Hugh Dickins58fa8792009-09-21 17:03:31 -0700160 * If FOLL_FORCE is set, we only require the "MAY" flags.
David Howells7b4d5b82006-09-27 01:50:18 -0700161 */
Hugh Dickins58fa8792009-09-21 17:03:31 -0700162 vm_flags = (foll_flags & FOLL_WRITE) ?
163 (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
164 vm_flags &= (foll_flags & FOLL_FORCE) ?
165 (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166
Peter Zijlstra9d737772009-06-25 11:58:55 +0200167 for (i = 0; i < nr_pages; i++) {
David Howells7561e8c2010-03-25 16:48:38 +0000168 vma = find_vma(mm, start);
David Howells7b4d5b82006-09-27 01:50:18 -0700169 if (!vma)
170 goto finish_or_fault;
171
172 /* protect what we can, including chardevs */
Hugh Dickins1c3aff12009-09-21 17:03:24 -0700173 if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
174 !(vm_flags & vma->vm_flags))
David Howells7b4d5b82006-09-27 01:50:18 -0700175 goto finish_or_fault;
Sonic Zhang910e46d2006-09-27 01:50:17 -0700176
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 if (pages) {
178 pages[i] = virt_to_page(start);
179 if (pages[i])
180 page_cache_get(pages[i]);
181 }
182 if (vmas)
Sonic Zhang910e46d2006-09-27 01:50:17 -0700183 vmas[i] = vma;
David Howellse1ee65d2010-03-25 16:48:44 +0000184 start = (start + PAGE_SIZE) & PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 }
David Howells7b4d5b82006-09-27 01:50:18 -0700186
187 return i;
188
189finish_or_fault:
190 return i ? : -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191}
Nick Pigginb291f002008-10-18 20:26:44 -0700192
Nick Pigginb291f002008-10-18 20:26:44 -0700193/*
194 * get a list of pages in an address range belonging to the specified process
195 * and indicate the VMA that covers each page
196 * - this is potentially dodgy as we may end incrementing the page count of a
197 * slab page or a secondary page from a compound page
198 * - don't permit access to VMAs that don't support it, such as I/O mappings
199 */
Michel Lespinasse28a35712013-02-22 16:35:55 -0800200long get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
201 unsigned long start, unsigned long nr_pages,
202 int write, int force, struct page **pages,
203 struct vm_area_struct **vmas)
Nick Pigginb291f002008-10-18 20:26:44 -0700204{
205 int flags = 0;
206
207 if (write)
Hugh Dickins58fa8792009-09-21 17:03:31 -0700208 flags |= FOLL_WRITE;
Nick Pigginb291f002008-10-18 20:26:44 -0700209 if (force)
Hugh Dickins58fa8792009-09-21 17:03:31 -0700210 flags |= FOLL_FORCE;
Nick Pigginb291f002008-10-18 20:26:44 -0700211
Michel Lespinasse53a77062011-01-13 15:46:14 -0800212 return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas,
213 NULL);
Nick Pigginb291f002008-10-18 20:26:44 -0700214}
Greg Ungerer66aa2b42005-09-12 11:18:10 +1000215EXPORT_SYMBOL(get_user_pages);
216
Andrea Arcangelif0818f42015-02-11 15:27:17 -0800217long get_user_pages_locked(struct task_struct *tsk, struct mm_struct *mm,
218 unsigned long start, unsigned long nr_pages,
219 int write, int force, struct page **pages,
220 int *locked)
221{
222 return get_user_pages(tsk, mm, start, nr_pages, write, force,
223 pages, NULL);
224}
225EXPORT_SYMBOL(get_user_pages_locked);
226
Andrea Arcangeli0fd71a52015-02-11 15:27:20 -0800227long __get_user_pages_unlocked(struct task_struct *tsk, struct mm_struct *mm,
228 unsigned long start, unsigned long nr_pages,
229 int write, int force, struct page **pages,
230 unsigned int gup_flags)
Andrea Arcangelif0818f42015-02-11 15:27:17 -0800231{
232 long ret;
233 down_read(&mm->mmap_sem);
234 ret = get_user_pages(tsk, mm, start, nr_pages, write, force,
235 pages, NULL);
236 up_read(&mm->mmap_sem);
237 return ret;
238}
Andrea Arcangeli0fd71a52015-02-11 15:27:20 -0800239EXPORT_SYMBOL(__get_user_pages_unlocked);
240
241long get_user_pages_unlocked(struct task_struct *tsk, struct mm_struct *mm,
242 unsigned long start, unsigned long nr_pages,
243 int write, int force, struct page **pages)
244{
245 return __get_user_pages_unlocked(tsk, mm, start, nr_pages, write,
246 force, pages, 0);
247}
Andrea Arcangelif0818f42015-02-11 15:27:17 -0800248EXPORT_SYMBOL(get_user_pages_unlocked);
249
Paul Mundtdfc2f912009-06-26 04:31:57 +0900250/**
251 * follow_pfn - look up PFN at a user virtual address
252 * @vma: memory mapping
253 * @address: user virtual address
254 * @pfn: location to store found PFN
255 *
256 * Only IO mappings and raw PFN mappings are allowed.
257 *
258 * Returns zero and the pfn at @pfn on success, -ve otherwise.
259 */
260int follow_pfn(struct vm_area_struct *vma, unsigned long address,
261 unsigned long *pfn)
262{
263 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
264 return -EINVAL;
265
266 *pfn = address >> PAGE_SHIFT;
267 return 0;
268}
269EXPORT_SYMBOL(follow_pfn);
270
Joonsoo Kimf1c40692013-04-29 15:07:37 -0700271LIST_HEAD(vmap_area_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800273void vfree(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274{
275 kfree(addr);
276}
Paul Mundtb5073172007-07-21 04:37:25 -0700277EXPORT_SYMBOL(vfree);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278
Al Virodd0fc662005-10-07 07:46:04 +0100279void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280{
281 /*
Robert P. J. Day85186092007-10-19 23:11:38 +0200282 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
283 * returns only a logical address.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 */
Nick Piggin84097512006-03-22 00:08:34 -0800285 return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286}
Paul Mundtb5073172007-07-21 04:37:25 -0700287EXPORT_SYMBOL(__vmalloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288
Paul Mundtf905bc42008-02-04 22:29:59 -0800289void *vmalloc_user(unsigned long size)
290{
291 void *ret;
292
293 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
294 PAGE_KERNEL);
295 if (ret) {
296 struct vm_area_struct *vma;
297
298 down_write(&current->mm->mmap_sem);
299 vma = find_vma(current->mm, (unsigned long)ret);
300 if (vma)
301 vma->vm_flags |= VM_USERMAP;
302 up_write(&current->mm->mmap_sem);
303 }
304
305 return ret;
306}
307EXPORT_SYMBOL(vmalloc_user);
308
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800309struct page *vmalloc_to_page(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310{
311 return virt_to_page(addr);
312}
Paul Mundtb5073172007-07-21 04:37:25 -0700313EXPORT_SYMBOL(vmalloc_to_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800315unsigned long vmalloc_to_pfn(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316{
317 return page_to_pfn(virt_to_page(addr));
318}
Paul Mundtb5073172007-07-21 04:37:25 -0700319EXPORT_SYMBOL(vmalloc_to_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320
321long vread(char *buf, char *addr, unsigned long count)
322{
Chen Gang9bde9162013-07-03 15:02:36 -0700323 /* Don't allow overflow */
324 if ((unsigned long) buf + count < count)
325 count = -(unsigned long) buf;
326
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 memcpy(buf, addr, count);
328 return count;
329}
330
331long vwrite(char *buf, char *addr, unsigned long count)
332{
333 /* Don't allow overflow */
334 if ((unsigned long) addr + count < count)
335 count = -(unsigned long) addr;
336
337 memcpy(addr, buf, count);
Choi Gi-yongac714902014-04-07 15:37:36 -0700338 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339}
340
341/*
342 * vmalloc - allocate virtually continguos memory
343 *
344 * @size: allocation size
345 *
346 * Allocate enough pages to cover @size from the page level
347 * allocator and map them into continguos kernel virtual space.
348 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +0200349 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 * use __vmalloc() instead.
351 */
352void *vmalloc(unsigned long size)
353{
354 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
355}
Andrew Mortonf6138882006-02-28 16:59:18 -0800356EXPORT_SYMBOL(vmalloc);
357
Dave Younge1ca7782010-10-26 14:22:06 -0700358/*
359 * vzalloc - allocate virtually continguos memory with zero fill
360 *
361 * @size: allocation size
362 *
363 * Allocate enough pages to cover @size from the page level
364 * allocator and map them into continguos kernel virtual space.
365 * The memory allocated is set to zero.
366 *
367 * For tight control over page level allocator and protection flags
368 * use __vmalloc() instead.
369 */
370void *vzalloc(unsigned long size)
371{
372 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
373 PAGE_KERNEL);
374}
375EXPORT_SYMBOL(vzalloc);
376
377/**
378 * vmalloc_node - allocate memory on a specific node
379 * @size: allocation size
380 * @node: numa node
381 *
382 * Allocate enough pages to cover @size from the page level
383 * allocator and map them into contiguous kernel virtual space.
384 *
385 * For tight control over page level allocator and protection flags
386 * use __vmalloc() instead.
387 */
Andrew Mortonf6138882006-02-28 16:59:18 -0800388void *vmalloc_node(unsigned long size, int node)
389{
390 return vmalloc(size);
391}
Paul Mundt9a14f652010-12-24 11:50:34 +0900392EXPORT_SYMBOL(vmalloc_node);
Dave Younge1ca7782010-10-26 14:22:06 -0700393
394/**
395 * vzalloc_node - allocate memory on a specific node with zero fill
396 * @size: allocation size
397 * @node: numa node
398 *
399 * Allocate enough pages to cover @size from the page level
400 * allocator and map them into contiguous kernel virtual space.
401 * The memory allocated is set to zero.
402 *
403 * For tight control over page level allocator and protection flags
404 * use __vmalloc() instead.
405 */
406void *vzalloc_node(unsigned long size, int node)
407{
408 return vzalloc(size);
409}
410EXPORT_SYMBOL(vzalloc_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411
Paul Mundt1af446e2008-08-04 16:01:47 +0900412#ifndef PAGE_KERNEL_EXEC
413# define PAGE_KERNEL_EXEC PAGE_KERNEL
414#endif
415
416/**
417 * vmalloc_exec - allocate virtually contiguous, executable memory
418 * @size: allocation size
419 *
420 * Kernel-internal function to allocate enough pages to cover @size
421 * the page level allocator and map them into contiguous and
422 * executable kernel virtual space.
423 *
424 * For tight control over page level allocator and protection flags
425 * use __vmalloc() instead.
426 */
427
428void *vmalloc_exec(unsigned long size)
429{
430 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
431}
432
Paul Mundtb5073172007-07-21 04:37:25 -0700433/**
434 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 * @size: allocation size
436 *
437 * Allocate enough 32bit PA addressable pages to cover @size from the
438 * page level allocator and map them into continguos kernel virtual space.
439 */
440void *vmalloc_32(unsigned long size)
441{
442 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
443}
Paul Mundtb5073172007-07-21 04:37:25 -0700444EXPORT_SYMBOL(vmalloc_32);
445
446/**
447 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
448 * @size: allocation size
449 *
450 * The resulting memory area is 32bit addressable and zeroed so it can be
451 * mapped to userspace without leaking data.
Paul Mundtf905bc42008-02-04 22:29:59 -0800452 *
453 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
454 * remap_vmalloc_range() are permissible.
Paul Mundtb5073172007-07-21 04:37:25 -0700455 */
456void *vmalloc_32_user(unsigned long size)
457{
Paul Mundtf905bc42008-02-04 22:29:59 -0800458 /*
459 * We'll have to sort out the ZONE_DMA bits for 64-bit,
460 * but for now this can simply use vmalloc_user() directly.
461 */
462 return vmalloc_user(size);
Paul Mundtb5073172007-07-21 04:37:25 -0700463}
464EXPORT_SYMBOL(vmalloc_32_user);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465
466void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
467{
468 BUG();
469 return NULL;
470}
Paul Mundtb5073172007-07-21 04:37:25 -0700471EXPORT_SYMBOL(vmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800473void vunmap(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474{
475 BUG();
476}
Paul Mundtb5073172007-07-21 04:37:25 -0700477EXPORT_SYMBOL(vunmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478
Paul Mundteb6434d2009-01-21 17:45:47 +0900479void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
480{
481 BUG();
482 return NULL;
483}
484EXPORT_SYMBOL(vm_map_ram);
485
486void vm_unmap_ram(const void *mem, unsigned int count)
487{
488 BUG();
489}
490EXPORT_SYMBOL(vm_unmap_ram);
491
492void vm_unmap_aliases(void)
493{
494}
495EXPORT_SYMBOL_GPL(vm_unmap_aliases);
496
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497/*
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -0700498 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
499 * have one.
500 */
Gideon Israel Dsouza3b321232014-04-07 15:37:26 -0700501void __weak vmalloc_sync_all(void)
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -0700502{
503}
504
Paul Mundt29c185e2010-12-24 12:08:30 +0900505/**
506 * alloc_vm_area - allocate a range of kernel address space
507 * @size: size of the area
508 *
509 * Returns: NULL on failure, vm_struct on success
510 *
511 * This function reserves a range of kernel address space, and
512 * allocates pagetables to map that range. No actual mappings
513 * are created. If the kernel address space is not shared
514 * between processes, it syncs the pagetable across all
515 * processes.
516 */
David Vrabelcd129092011-09-29 16:53:32 +0100517struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
Paul Mundt29c185e2010-12-24 12:08:30 +0900518{
519 BUG();
520 return NULL;
521}
522EXPORT_SYMBOL_GPL(alloc_vm_area);
523
524void free_vm_area(struct vm_struct *area)
525{
526 BUG();
527}
528EXPORT_SYMBOL_GPL(free_vm_area);
529
Paul Mundtb5073172007-07-21 04:37:25 -0700530int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
531 struct page *page)
532{
533 return -EINVAL;
534}
535EXPORT_SYMBOL(vm_insert_page);
536
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -0700537/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 * sys_brk() for the most part doesn't need the global kernel
539 * lock, except when an application is doing something nasty
540 * like trying to un-brk an area that has already been mapped
541 * to a regular file. in this case, the unmapping will need
542 * to invoke file system routines that need the global lock.
543 */
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100544SYSCALL_DEFINE1(brk, unsigned long, brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545{
546 struct mm_struct *mm = current->mm;
547
548 if (brk < mm->start_brk || brk > mm->context.end_brk)
549 return mm->brk;
550
551 if (mm->brk == brk)
552 return mm->brk;
553
554 /*
555 * Always allow shrinking brk
556 */
557 if (brk <= mm->brk) {
558 mm->brk = brk;
559 return brk;
560 }
561
562 /*
563 * Ok, looks good - let it rip.
564 */
Mike Frysingercfe79c02010-01-06 17:23:23 +0000565 flush_icache_range(mm->brk, brk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 return mm->brk = brk;
567}
568
David Howells8feae132009-01-08 12:04:47 +0000569/*
570 * initialise the VMA and region record slabs
571 */
572void __init mmap_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573{
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -0700574 int ret;
575
Tejun Heo908c7f12014-09-08 09:51:29 +0900576 ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -0700577 VM_BUG_ON(ret);
David Howells33e5d7692009-04-02 16:56:32 -0700578 vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC);
David Howells8feae132009-01-08 12:04:47 +0000579}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580
David Howells8feae132009-01-08 12:04:47 +0000581/*
582 * validate the region tree
583 * - the caller must hold the region lock
584 */
585#ifdef CONFIG_DEBUG_NOMMU_REGIONS
586static noinline void validate_nommu_regions(void)
587{
588 struct vm_region *region, *last;
589 struct rb_node *p, *lastp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590
David Howells8feae132009-01-08 12:04:47 +0000591 lastp = rb_first(&nommu_region_tree);
592 if (!lastp)
593 return;
594
595 last = rb_entry(lastp, struct vm_region, vm_rb);
David Howells33e5d7692009-04-02 16:56:32 -0700596 BUG_ON(unlikely(last->vm_end <= last->vm_start));
597 BUG_ON(unlikely(last->vm_top < last->vm_end));
David Howells8feae132009-01-08 12:04:47 +0000598
599 while ((p = rb_next(lastp))) {
600 region = rb_entry(p, struct vm_region, vm_rb);
601 last = rb_entry(lastp, struct vm_region, vm_rb);
602
David Howells33e5d7692009-04-02 16:56:32 -0700603 BUG_ON(unlikely(region->vm_end <= region->vm_start));
604 BUG_ON(unlikely(region->vm_top < region->vm_end));
605 BUG_ON(unlikely(region->vm_start < last->vm_top));
David Howells8feae132009-01-08 12:04:47 +0000606
607 lastp = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 }
609}
David Howells8feae132009-01-08 12:04:47 +0000610#else
David Howells33e5d7692009-04-02 16:56:32 -0700611static void validate_nommu_regions(void)
612{
613}
David Howells8feae132009-01-08 12:04:47 +0000614#endif
615
616/*
617 * add a region into the global tree
618 */
619static void add_nommu_region(struct vm_region *region)
620{
621 struct vm_region *pregion;
622 struct rb_node **p, *parent;
623
624 validate_nommu_regions();
625
David Howells8feae132009-01-08 12:04:47 +0000626 parent = NULL;
627 p = &nommu_region_tree.rb_node;
628 while (*p) {
629 parent = *p;
630 pregion = rb_entry(parent, struct vm_region, vm_rb);
631 if (region->vm_start < pregion->vm_start)
632 p = &(*p)->rb_left;
633 else if (region->vm_start > pregion->vm_start)
634 p = &(*p)->rb_right;
635 else if (pregion == region)
636 return;
637 else
638 BUG();
639 }
640
641 rb_link_node(&region->vm_rb, parent, p);
642 rb_insert_color(&region->vm_rb, &nommu_region_tree);
643
644 validate_nommu_regions();
645}
646
647/*
648 * delete a region from the global tree
649 */
650static void delete_nommu_region(struct vm_region *region)
651{
652 BUG_ON(!nommu_region_tree.rb_node);
653
654 validate_nommu_regions();
655 rb_erase(&region->vm_rb, &nommu_region_tree);
656 validate_nommu_regions();
657}
658
659/*
660 * free a contiguous series of pages
661 */
662static void free_page_series(unsigned long from, unsigned long to)
663{
664 for (; from < to; from += PAGE_SIZE) {
665 struct page *page = virt_to_page(from);
666
667 kdebug("- free %lx", from);
David Howells33e5d7692009-04-02 16:56:32 -0700668 atomic_long_dec(&mmap_pages_allocated);
David Howells8feae132009-01-08 12:04:47 +0000669 if (page_count(page) != 1)
David Howells33e5d7692009-04-02 16:56:32 -0700670 kdebug("free page %p: refcount not one: %d",
671 page, page_count(page));
David Howells8feae132009-01-08 12:04:47 +0000672 put_page(page);
673 }
674}
675
676/*
677 * release a reference to a region
David Howells33e5d7692009-04-02 16:56:32 -0700678 * - the caller must hold the region semaphore for writing, which this releases
Paul Mundtdd8632a2009-01-08 12:04:47 +0000679 * - the region may not have been added to the tree yet, in which case vm_top
David Howells8feae132009-01-08 12:04:47 +0000680 * will equal vm_start
681 */
682static void __put_nommu_region(struct vm_region *region)
683 __releases(nommu_region_sem)
684{
David Howells1e2ae592010-01-15 17:01:33 -0800685 kenter("%p{%d}", region, region->vm_usage);
David Howells8feae132009-01-08 12:04:47 +0000686
687 BUG_ON(!nommu_region_tree.rb_node);
688
David Howells1e2ae592010-01-15 17:01:33 -0800689 if (--region->vm_usage == 0) {
Paul Mundtdd8632a2009-01-08 12:04:47 +0000690 if (region->vm_top > region->vm_start)
David Howells8feae132009-01-08 12:04:47 +0000691 delete_nommu_region(region);
692 up_write(&nommu_region_sem);
693
694 if (region->vm_file)
695 fput(region->vm_file);
696
697 /* IO memory and memory shared directly out of the pagecache
698 * from ramfs/tmpfs mustn't be released here */
699 if (region->vm_flags & VM_MAPPED_COPY) {
700 kdebug("free series");
Paul Mundtdd8632a2009-01-08 12:04:47 +0000701 free_page_series(region->vm_start, region->vm_top);
David Howells8feae132009-01-08 12:04:47 +0000702 }
703 kmem_cache_free(vm_region_jar, region);
704 } else {
705 up_write(&nommu_region_sem);
706 }
707}
708
709/*
710 * release a reference to a region
711 */
712static void put_nommu_region(struct vm_region *region)
713{
714 down_write(&nommu_region_sem);
715 __put_nommu_region(region);
716}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717
David Howells30340972006-09-27 01:50:20 -0700718/*
Bernd Schmidteb8cdec2009-09-21 17:03:57 -0700719 * update protection on a vma
720 */
721static void protect_vma(struct vm_area_struct *vma, unsigned long flags)
722{
723#ifdef CONFIG_MPU
724 struct mm_struct *mm = vma->vm_mm;
725 long start = vma->vm_start & PAGE_MASK;
726 while (start < vma->vm_end) {
727 protect_page(mm, start, flags);
728 start += PAGE_SIZE;
729 }
730 update_protections(mm);
731#endif
732}
733
734/*
David Howells30340972006-09-27 01:50:20 -0700735 * add a VMA into a process's mm_struct in the appropriate place in the list
David Howells8feae132009-01-08 12:04:47 +0000736 * and tree and add to the address space's page tree also if not an anonymous
737 * page
David Howells30340972006-09-27 01:50:20 -0700738 * - should be called with mm->mmap_sem held writelocked
739 */
David Howells8feae132009-01-08 12:04:47 +0000740static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
David Howells30340972006-09-27 01:50:20 -0700741{
Namhyung Kim6038def2011-05-24 17:11:22 -0700742 struct vm_area_struct *pvma, *prev;
David Howells8feae132009-01-08 12:04:47 +0000743 struct address_space *mapping;
Namhyung Kim6038def2011-05-24 17:11:22 -0700744 struct rb_node **p, *parent, *rb_prev;
David Howells30340972006-09-27 01:50:20 -0700745
David Howells8feae132009-01-08 12:04:47 +0000746 kenter(",%p", vma);
747
748 BUG_ON(!vma->vm_region);
749
750 mm->map_count++;
751 vma->vm_mm = mm;
752
Bernd Schmidteb8cdec2009-09-21 17:03:57 -0700753 protect_vma(vma, vma->vm_flags);
754
David Howells8feae132009-01-08 12:04:47 +0000755 /* add the VMA to the mapping */
756 if (vma->vm_file) {
757 mapping = vma->vm_file->f_mapping;
758
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800759 i_mmap_lock_write(mapping);
David Howells8feae132009-01-08 12:04:47 +0000760 flush_dcache_mmap_lock(mapping);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700761 vma_interval_tree_insert(vma, &mapping->i_mmap);
David Howells8feae132009-01-08 12:04:47 +0000762 flush_dcache_mmap_unlock(mapping);
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800763 i_mmap_unlock_write(mapping);
David Howells8feae132009-01-08 12:04:47 +0000764 }
765
766 /* add the VMA to the tree */
Namhyung Kim6038def2011-05-24 17:11:22 -0700767 parent = rb_prev = NULL;
David Howells8feae132009-01-08 12:04:47 +0000768 p = &mm->mm_rb.rb_node;
769 while (*p) {
770 parent = *p;
771 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
772
773 /* sort by: start addr, end addr, VMA struct addr in that order
774 * (the latter is necessary as we may get identical VMAs) */
775 if (vma->vm_start < pvma->vm_start)
776 p = &(*p)->rb_left;
Namhyung Kim6038def2011-05-24 17:11:22 -0700777 else if (vma->vm_start > pvma->vm_start) {
778 rb_prev = parent;
David Howells8feae132009-01-08 12:04:47 +0000779 p = &(*p)->rb_right;
Namhyung Kim6038def2011-05-24 17:11:22 -0700780 } else if (vma->vm_end < pvma->vm_end)
David Howells8feae132009-01-08 12:04:47 +0000781 p = &(*p)->rb_left;
Namhyung Kim6038def2011-05-24 17:11:22 -0700782 else if (vma->vm_end > pvma->vm_end) {
783 rb_prev = parent;
David Howells8feae132009-01-08 12:04:47 +0000784 p = &(*p)->rb_right;
Namhyung Kim6038def2011-05-24 17:11:22 -0700785 } else if (vma < pvma)
David Howells8feae132009-01-08 12:04:47 +0000786 p = &(*p)->rb_left;
Namhyung Kim6038def2011-05-24 17:11:22 -0700787 else if (vma > pvma) {
788 rb_prev = parent;
David Howells8feae132009-01-08 12:04:47 +0000789 p = &(*p)->rb_right;
Namhyung Kim6038def2011-05-24 17:11:22 -0700790 } else
David Howells8feae132009-01-08 12:04:47 +0000791 BUG();
792 }
793
794 rb_link_node(&vma->vm_rb, parent, p);
795 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
796
797 /* add VMA to the VMA list also */
Namhyung Kim6038def2011-05-24 17:11:22 -0700798 prev = NULL;
799 if (rb_prev)
800 prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
David Howells30340972006-09-27 01:50:20 -0700801
Namhyung Kim6038def2011-05-24 17:11:22 -0700802 __vma_link_list(mm, vma, prev, parent);
David Howells8feae132009-01-08 12:04:47 +0000803}
804
805/*
806 * delete a VMA from its owning mm_struct and address space
807 */
808static void delete_vma_from_mm(struct vm_area_struct *vma)
809{
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700810 int i;
David Howells8feae132009-01-08 12:04:47 +0000811 struct address_space *mapping;
812 struct mm_struct *mm = vma->vm_mm;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700813 struct task_struct *curr = current;
David Howells8feae132009-01-08 12:04:47 +0000814
815 kenter("%p", vma);
816
Bernd Schmidteb8cdec2009-09-21 17:03:57 -0700817 protect_vma(vma, 0);
818
David Howells8feae132009-01-08 12:04:47 +0000819 mm->map_count--;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700820 for (i = 0; i < VMACACHE_SIZE; i++) {
821 /* if the vma is cached, invalidate the entire cache */
822 if (curr->vmacache[i] == vma) {
Steven Miaoe020d5b2014-06-23 13:22:02 -0700823 vmacache_invalidate(mm);
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700824 break;
825 }
826 }
David Howells8feae132009-01-08 12:04:47 +0000827
828 /* remove the VMA from the mapping */
829 if (vma->vm_file) {
830 mapping = vma->vm_file->f_mapping;
831
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800832 i_mmap_lock_write(mapping);
David Howells8feae132009-01-08 12:04:47 +0000833 flush_dcache_mmap_lock(mapping);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700834 vma_interval_tree_remove(vma, &mapping->i_mmap);
David Howells8feae132009-01-08 12:04:47 +0000835 flush_dcache_mmap_unlock(mapping);
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800836 i_mmap_unlock_write(mapping);
David Howells8feae132009-01-08 12:04:47 +0000837 }
838
839 /* remove from the MM's tree and list */
840 rb_erase(&vma->vm_rb, &mm->mm_rb);
Namhyung Kimb951bf22011-05-24 17:11:23 -0700841
842 if (vma->vm_prev)
843 vma->vm_prev->vm_next = vma->vm_next;
844 else
845 mm->mmap = vma->vm_next;
846
847 if (vma->vm_next)
848 vma->vm_next->vm_prev = vma->vm_prev;
David Howells8feae132009-01-08 12:04:47 +0000849}
850
851/*
852 * destroy a VMA record
853 */
854static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
855{
856 kenter("%p", vma);
857 if (vma->vm_ops && vma->vm_ops->close)
858 vma->vm_ops->close(vma);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -0700859 if (vma->vm_file)
David Howells8feae132009-01-08 12:04:47 +0000860 fput(vma->vm_file);
David Howells8feae132009-01-08 12:04:47 +0000861 put_nommu_region(vma->vm_region);
862 kmem_cache_free(vm_area_cachep, vma);
David Howells30340972006-09-27 01:50:20 -0700863}
864
865/*
866 * look up the first VMA in which addr resides, NULL if none
867 * - should be called with mm->mmap_sem at least held readlocked
868 */
869struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
870{
David Howells8feae132009-01-08 12:04:47 +0000871 struct vm_area_struct *vma;
David Howells30340972006-09-27 01:50:20 -0700872
David Howells8feae132009-01-08 12:04:47 +0000873 /* check the cache first */
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700874 vma = vmacache_find(mm, addr);
875 if (likely(vma))
David Howells8feae132009-01-08 12:04:47 +0000876 return vma;
877
Namhyung Kime922c4c2011-05-24 17:11:24 -0700878 /* trawl the list (there may be multiple mappings in which addr
David Howells8feae132009-01-08 12:04:47 +0000879 * resides) */
Namhyung Kime922c4c2011-05-24 17:11:24 -0700880 for (vma = mm->mmap; vma; vma = vma->vm_next) {
David Howells8feae132009-01-08 12:04:47 +0000881 if (vma->vm_start > addr)
882 return NULL;
883 if (vma->vm_end > addr) {
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700884 vmacache_update(addr, vma);
David Howells8feae132009-01-08 12:04:47 +0000885 return vma;
886 }
David Howells30340972006-09-27 01:50:20 -0700887 }
888
David Howells30340972006-09-27 01:50:20 -0700889 return NULL;
890}
891EXPORT_SYMBOL(find_vma);
892
893/*
David Howells930e6522006-09-27 01:50:22 -0700894 * find a VMA
895 * - we don't extend stack VMAs under NOMMU conditions
896 */
897struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
898{
David Howells7561e8c2010-03-25 16:48:38 +0000899 return find_vma(mm, addr);
David Howells930e6522006-09-27 01:50:22 -0700900}
901
David Howells8feae132009-01-08 12:04:47 +0000902/*
903 * expand a stack to a given address
904 * - not supported under NOMMU conditions
905 */
Greg Ungerer57c8f632007-07-15 23:38:28 -0700906int expand_stack(struct vm_area_struct *vma, unsigned long address)
907{
908 return -ENOMEM;
909}
910
David Howells930e6522006-09-27 01:50:22 -0700911/*
David Howells6fa5f802006-09-27 01:50:21 -0700912 * look up the first VMA exactly that exactly matches addr
913 * - should be called with mm->mmap_sem at least held readlocked
914 */
David Howells8feae132009-01-08 12:04:47 +0000915static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
916 unsigned long addr,
917 unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918{
919 struct vm_area_struct *vma;
David Howells8feae132009-01-08 12:04:47 +0000920 unsigned long end = addr + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921
David Howells8feae132009-01-08 12:04:47 +0000922 /* check the cache first */
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700923 vma = vmacache_find_exact(mm, addr, end);
924 if (vma)
David Howells8feae132009-01-08 12:04:47 +0000925 return vma;
926
Namhyung Kime922c4c2011-05-24 17:11:24 -0700927 /* trawl the list (there may be multiple mappings in which addr
David Howells8feae132009-01-08 12:04:47 +0000928 * resides) */
Namhyung Kime922c4c2011-05-24 17:11:24 -0700929 for (vma = mm->mmap; vma; vma = vma->vm_next) {
David Howells8feae132009-01-08 12:04:47 +0000930 if (vma->vm_start < addr)
931 continue;
932 if (vma->vm_start > addr)
933 return NULL;
934 if (vma->vm_end == end) {
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700935 vmacache_update(addr, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 return vma;
David Howells8feae132009-01-08 12:04:47 +0000937 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938 }
939
940 return NULL;
941}
942
David Howells30340972006-09-27 01:50:20 -0700943/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 * determine whether a mapping should be permitted and, if so, what sort of
945 * mapping we're capable of supporting
946 */
947static int validate_mmap_request(struct file *file,
948 unsigned long addr,
949 unsigned long len,
950 unsigned long prot,
951 unsigned long flags,
952 unsigned long pgoff,
953 unsigned long *_capabilities)
954{
David Howells8feae132009-01-08 12:04:47 +0000955 unsigned long capabilities, rlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 int ret;
957
958 /* do the simple checks first */
David Howells06aab5a2009-09-24 12:33:48 +0100959 if (flags & MAP_FIXED) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 printk(KERN_DEBUG
961 "%d: Can't do fixed-address/overlay mmap of RAM\n",
962 current->pid);
963 return -EINVAL;
964 }
965
966 if ((flags & MAP_TYPE) != MAP_PRIVATE &&
967 (flags & MAP_TYPE) != MAP_SHARED)
968 return -EINVAL;
969
Mike Frysingerf81cff02006-12-06 12:02:59 +1000970 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 return -EINVAL;
972
Mike Frysingerf81cff02006-12-06 12:02:59 +1000973 /* Careful about overflows.. */
David Howells8feae132009-01-08 12:04:47 +0000974 rlen = PAGE_ALIGN(len);
975 if (!rlen || rlen > TASK_SIZE)
Mike Frysingerf81cff02006-12-06 12:02:59 +1000976 return -ENOMEM;
977
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 /* offset overflow? */
David Howells8feae132009-01-08 12:04:47 +0000979 if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
Mike Frysingerf81cff02006-12-06 12:02:59 +1000980 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981
982 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 /* files must support mmap */
Al Viro72c2d532013-09-22 16:27:52 -0400984 if (!file->f_op->mmap)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 return -ENODEV;
986
987 /* work out if what we've got could possibly be shared
988 * - we support chardevs that provide their own "memory"
989 * - we support files/blockdevs that are memory backed
990 */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100991 if (file->f_op->mmap_capabilities) {
992 capabilities = file->f_op->mmap_capabilities(file);
993 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 /* no explicit capabilities set, so assume some
995 * defaults */
Al Viro496ad9a2013-01-23 17:07:38 -0500996 switch (file_inode(file)->i_mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 case S_IFREG:
998 case S_IFBLK:
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100999 capabilities = NOMMU_MAP_COPY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 break;
1001
1002 case S_IFCHR:
1003 capabilities =
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001004 NOMMU_MAP_DIRECT |
1005 NOMMU_MAP_READ |
1006 NOMMU_MAP_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 break;
1008
1009 default:
1010 return -EINVAL;
1011 }
1012 }
1013
1014 /* eliminate any capabilities that we can't support on this
1015 * device */
1016 if (!file->f_op->get_unmapped_area)
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001017 capabilities &= ~NOMMU_MAP_DIRECT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 if (!file->f_op->read)
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001019 capabilities &= ~NOMMU_MAP_COPY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020
Graff Yang28d7a6a2009-08-18 14:11:17 -07001021 /* The file shall have been opened with read permission. */
1022 if (!(file->f_mode & FMODE_READ))
1023 return -EACCES;
1024
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 if (flags & MAP_SHARED) {
1026 /* do checks for writing, appending and locking */
1027 if ((prot & PROT_WRITE) &&
1028 !(file->f_mode & FMODE_WRITE))
1029 return -EACCES;
1030
Al Viro496ad9a2013-01-23 17:07:38 -05001031 if (IS_APPEND(file_inode(file)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 (file->f_mode & FMODE_WRITE))
1033 return -EACCES;
1034
Jeff Laytond7a06982014-03-10 09:54:15 -04001035 if (locks_verify_locked(file))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 return -EAGAIN;
1037
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001038 if (!(capabilities & NOMMU_MAP_DIRECT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 return -ENODEV;
1040
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 /* we mustn't privatise shared mappings */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001042 capabilities &= ~NOMMU_MAP_COPY;
Choi Gi-yongac714902014-04-07 15:37:36 -07001043 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 /* we're going to read the file into private memory we
1045 * allocate */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001046 if (!(capabilities & NOMMU_MAP_COPY))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 return -ENODEV;
1048
1049 /* we don't permit a private writable mapping to be
1050 * shared with the backing device */
1051 if (prot & PROT_WRITE)
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001052 capabilities &= ~NOMMU_MAP_DIRECT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 }
1054
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001055 if (capabilities & NOMMU_MAP_DIRECT) {
1056 if (((prot & PROT_READ) && !(capabilities & NOMMU_MAP_READ)) ||
1057 ((prot & PROT_WRITE) && !(capabilities & NOMMU_MAP_WRITE)) ||
1058 ((prot & PROT_EXEC) && !(capabilities & NOMMU_MAP_EXEC))
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001059 ) {
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001060 capabilities &= ~NOMMU_MAP_DIRECT;
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001061 if (flags & MAP_SHARED) {
1062 printk(KERN_WARNING
1063 "MAP_SHARED not completely supported on !MMU\n");
1064 return -EINVAL;
1065 }
1066 }
1067 }
1068
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069 /* handle executable mappings and implied executable
1070 * mappings */
Josef Sipeke9536ae2006-12-08 02:37:21 -08001071 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 if (prot & PROT_EXEC)
1073 return -EPERM;
Choi Gi-yongac714902014-04-07 15:37:36 -07001074 } else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 /* handle implication of PROT_EXEC by PROT_READ */
1076 if (current->personality & READ_IMPLIES_EXEC) {
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001077 if (capabilities & NOMMU_MAP_EXEC)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 prot |= PROT_EXEC;
1079 }
Choi Gi-yongac714902014-04-07 15:37:36 -07001080 } else if ((prot & PROT_READ) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 (prot & PROT_EXEC) &&
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001082 !(capabilities & NOMMU_MAP_EXEC)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 ) {
1084 /* backing file is not executable, try to copy */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001085 capabilities &= ~NOMMU_MAP_DIRECT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 }
Choi Gi-yongac714902014-04-07 15:37:36 -07001087 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 /* anonymous mappings are always memory backed and can be
1089 * privately mapped
1090 */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001091 capabilities = NOMMU_MAP_COPY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092
1093 /* handle PROT_EXEC implication by PROT_READ */
1094 if ((prot & PROT_READ) &&
1095 (current->personality & READ_IMPLIES_EXEC))
1096 prot |= PROT_EXEC;
1097 }
1098
1099 /* allow the security API to have its say */
Al Viroe5467852012-05-30 13:30:51 -04001100 ret = security_mmap_addr(addr);
1101 if (ret < 0)
1102 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103
1104 /* looks okay */
1105 *_capabilities = capabilities;
1106 return 0;
1107}
1108
1109/*
1110 * we've determined that we can make the mapping, now translate what we
1111 * now know into VMA flags
1112 */
1113static unsigned long determine_vm_flags(struct file *file,
1114 unsigned long prot,
1115 unsigned long flags,
1116 unsigned long capabilities)
1117{
1118 unsigned long vm_flags;
1119
1120 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 /* vm_flags |= mm->def_flags; */
1122
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001123 if (!(capabilities & NOMMU_MAP_DIRECT)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 /* attempt to share read-only copies of mapped file chunks */
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001125 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 if (file && !(prot & PROT_WRITE))
1127 vm_flags |= VM_MAYSHARE;
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001128 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 /* overlay a shareable mapping on the backing device or inode
1130 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1131 * romfs/cramfs */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001132 vm_flags |= VM_MAYSHARE | (capabilities & NOMMU_VMFLAGS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 if (flags & MAP_SHARED)
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001134 vm_flags |= VM_SHARED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 }
1136
1137 /* refuse to let anyone share private mappings with this process if
1138 * it's being traced - otherwise breakpoints set in it may interfere
1139 * with another untraced process
1140 */
Tejun Heoa288eec2011-06-17 16:50:37 +02001141 if ((flags & MAP_PRIVATE) && current->ptrace)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142 vm_flags &= ~VM_MAYSHARE;
1143
1144 return vm_flags;
1145}
1146
1147/*
David Howells8feae132009-01-08 12:04:47 +00001148 * set up a shared mapping on a file (the driver or filesystem provides and
1149 * pins the storage)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 */
David Howells8feae132009-01-08 12:04:47 +00001151static int do_mmap_shared_file(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152{
1153 int ret;
1154
1155 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001156 if (ret == 0) {
1157 vma->vm_region->vm_top = vma->vm_region->vm_end;
David Howells645d83c2009-09-24 15:13:10 +01001158 return 0;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001159 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 if (ret != -ENOSYS)
1161 return ret;
1162
David Howells3fa30462010-03-23 13:35:21 -07001163 /* getting -ENOSYS indicates that direct mmap isn't possible (as
1164 * opposed to tried but failed) so we can only give a suitable error as
1165 * it's not possible to make a private copy if MAP_SHARED was given */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 return -ENODEV;
1167}
1168
1169/*
1170 * set up a private mapping or an anonymous shared mapping
1171 */
David Howells8feae132009-01-08 12:04:47 +00001172static int do_mmap_private(struct vm_area_struct *vma,
1173 struct vm_region *region,
David Howells645d83c2009-09-24 15:13:10 +01001174 unsigned long len,
1175 unsigned long capabilities)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176{
Joonsoo Kimdbc83582014-12-12 16:55:55 -08001177 unsigned long total, point;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 void *base;
David Howells8feae132009-01-08 12:04:47 +00001179 int ret, order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180
1181 /* invoke the file's mapping function so that it can keep track of
1182 * shared mappings on devices or memory
1183 * - VM_MAYSHARE will be set if it may attempt to share
1184 */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001185 if (capabilities & NOMMU_MAP_DIRECT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001187 if (ret == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 /* shouldn't return success if we're not sharing */
Paul Mundtdd8632a2009-01-08 12:04:47 +00001189 BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1190 vma->vm_region->vm_top = vma->vm_region->vm_end;
David Howells645d83c2009-09-24 15:13:10 +01001191 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 }
Paul Mundtdd8632a2009-01-08 12:04:47 +00001193 if (ret != -ENOSYS)
1194 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195
1196 /* getting an ENOSYS error indicates that direct mmap isn't
1197 * possible (as opposed to tried but failed) so we'll try to
1198 * make a private copy of the data and map that instead */
1199 }
1200
David Howells8feae132009-01-08 12:04:47 +00001201
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 /* allocate some memory to hold the mapping
1203 * - note that this may not return a page-aligned address if the object
1204 * we're allocating is smaller than a page
1205 */
Bob Liuf67d9b12011-05-24 17:12:56 -07001206 order = get_order(len);
David Howells8feae132009-01-08 12:04:47 +00001207 kdebug("alloc order %d for %lx", order, len);
1208
David Howells8feae132009-01-08 12:04:47 +00001209 total = 1 << order;
Bob Liuf67d9b12011-05-24 17:12:56 -07001210 point = len >> PAGE_SHIFT;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001211
Joonsoo Kimdbc83582014-12-12 16:55:55 -08001212 /* we don't want to allocate a power-of-2 sized page set */
Paul Mundtdd8632a2009-01-08 12:04:47 +00001213 if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) {
Joonsoo Kimdbc83582014-12-12 16:55:55 -08001214 total = point;
1215 kdebug("try to alloc exact %lu pages", total);
1216 base = alloc_pages_exact(len, GFP_KERNEL);
1217 } else {
1218 base = (void *)__get_free_pages(GFP_KERNEL, order);
David Howells8feae132009-01-08 12:04:47 +00001219 }
1220
Joonsoo Kimdbc83582014-12-12 16:55:55 -08001221 if (!base)
1222 goto enomem;
David Howells8feae132009-01-08 12:04:47 +00001223
Joonsoo Kimdbc83582014-12-12 16:55:55 -08001224 atomic_long_add(total, &mmap_pages_allocated);
1225
David Howells8feae132009-01-08 12:04:47 +00001226 region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1227 region->vm_start = (unsigned long) base;
Bob Liuf67d9b12011-05-24 17:12:56 -07001228 region->vm_end = region->vm_start + len;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001229 region->vm_top = region->vm_start + (total << PAGE_SHIFT);
David Howells8feae132009-01-08 12:04:47 +00001230
1231 vma->vm_start = region->vm_start;
1232 vma->vm_end = region->vm_start + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233
1234 if (vma->vm_file) {
1235 /* read the contents of a file into the copy */
1236 mm_segment_t old_fs;
1237 loff_t fpos;
1238
1239 fpos = vma->vm_pgoff;
1240 fpos <<= PAGE_SHIFT;
1241
1242 old_fs = get_fs();
1243 set_fs(KERNEL_DS);
Bob Liuf67d9b12011-05-24 17:12:56 -07001244 ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 set_fs(old_fs);
1246
1247 if (ret < 0)
1248 goto error_free;
1249
1250 /* clear the last little bit */
Bob Liuf67d9b12011-05-24 17:12:56 -07001251 if (ret < len)
1252 memset(base + ret, 0, len - ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 }
1255
1256 return 0;
1257
1258error_free:
Namhyung Kim7223bb42011-05-24 17:11:26 -07001259 free_page_series(region->vm_start, region->vm_top);
David Howells8feae132009-01-08 12:04:47 +00001260 region->vm_start = vma->vm_start = 0;
1261 region->vm_end = vma->vm_end = 0;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001262 region->vm_top = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 return ret;
1264
1265enomem:
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -07001266 pr_err("Allocation of length %lu from process %d (%s) failed\n",
Greg Ungerer05ae6fa2009-01-13 17:30:22 +10001267 len, current->pid, current->comm);
David Rientjes7bf02ea2011-05-24 17:11:16 -07001268 show_free_areas(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 return -ENOMEM;
1270}
1271
1272/*
1273 * handle mapping creation for uClinux
1274 */
Al Viroe3fc6292012-05-30 20:08:42 -04001275unsigned long do_mmap_pgoff(struct file *file,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 unsigned long addr,
1277 unsigned long len,
1278 unsigned long prot,
1279 unsigned long flags,
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001280 unsigned long pgoff,
Michel Lespinasse41badc12013-02-22 16:32:47 -08001281 unsigned long *populate)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282{
David Howells8feae132009-01-08 12:04:47 +00001283 struct vm_area_struct *vma;
1284 struct vm_region *region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 struct rb_node *rb;
David Howells8feae132009-01-08 12:04:47 +00001286 unsigned long capabilities, vm_flags, result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 int ret;
1288
David Howells8feae132009-01-08 12:04:47 +00001289 kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff);
1290
Michel Lespinasse41badc12013-02-22 16:32:47 -08001291 *populate = 0;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001292
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 /* decide whether we should attempt the mapping, and if so what sort of
1294 * mapping */
1295 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1296 &capabilities);
David Howells8feae132009-01-08 12:04:47 +00001297 if (ret < 0) {
1298 kleave(" = %d [val]", ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 return ret;
David Howells8feae132009-01-08 12:04:47 +00001300 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301
David Howells06aab5a2009-09-24 12:33:48 +01001302 /* we ignore the address hint */
1303 addr = 0;
Bob Liuf67d9b12011-05-24 17:12:56 -07001304 len = PAGE_ALIGN(len);
David Howells06aab5a2009-09-24 12:33:48 +01001305
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 /* we've determined that we can make the mapping, now translate what we
1307 * now know into VMA flags */
1308 vm_flags = determine_vm_flags(file, prot, flags, capabilities);
1309
David Howells8feae132009-01-08 12:04:47 +00001310 /* we're going to need to record the mapping */
1311 region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1312 if (!region)
1313 goto error_getting_region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314
David Howells8feae132009-01-08 12:04:47 +00001315 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1316 if (!vma)
1317 goto error_getting_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318
David Howells1e2ae592010-01-15 17:01:33 -08001319 region->vm_usage = 1;
David Howells8feae132009-01-08 12:04:47 +00001320 region->vm_flags = vm_flags;
1321 region->vm_pgoff = pgoff;
1322
Rik van Riel5beb4932010-03-05 13:42:07 -08001323 INIT_LIST_HEAD(&vma->anon_vma_chain);
David Howells8feae132009-01-08 12:04:47 +00001324 vma->vm_flags = vm_flags;
1325 vma->vm_pgoff = pgoff;
1326
1327 if (file) {
Al Virocb0942b2012-08-27 14:48:26 -04001328 region->vm_file = get_file(file);
1329 vma->vm_file = get_file(file);
David Howells8feae132009-01-08 12:04:47 +00001330 }
1331
1332 down_write(&nommu_region_sem);
1333
1334 /* if we want to share, we need to check for regions created by other
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 * mmap() calls that overlap with our proposed mapping
David Howells8feae132009-01-08 12:04:47 +00001336 * - we can only share with a superset match on most regular files
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 * - shared mappings on character devices and memory backed files are
1338 * permitted to overlap inexactly as far as we are concerned for in
1339 * these cases, sharing is handled in the driver or filesystem rather
1340 * than here
1341 */
1342 if (vm_flags & VM_MAYSHARE) {
David Howells8feae132009-01-08 12:04:47 +00001343 struct vm_region *pregion;
1344 unsigned long pglen, rpglen, pgend, rpgend, start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345
David Howells8feae132009-01-08 12:04:47 +00001346 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1347 pgend = pgoff + pglen;
David Howells165b2392007-03-22 00:11:24 -08001348
David Howells8feae132009-01-08 12:04:47 +00001349 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1350 pregion = rb_entry(rb, struct vm_region, vm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351
David Howells8feae132009-01-08 12:04:47 +00001352 if (!(pregion->vm_flags & VM_MAYSHARE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353 continue;
1354
1355 /* search for overlapping mappings on the same file */
Al Viro496ad9a2013-01-23 17:07:38 -05001356 if (file_inode(pregion->vm_file) !=
1357 file_inode(file))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358 continue;
1359
David Howells8feae132009-01-08 12:04:47 +00001360 if (pregion->vm_pgoff >= pgend)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 continue;
1362
David Howells8feae132009-01-08 12:04:47 +00001363 rpglen = pregion->vm_end - pregion->vm_start;
1364 rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1365 rpgend = pregion->vm_pgoff + rpglen;
1366 if (pgoff >= rpgend)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367 continue;
1368
David Howells8feae132009-01-08 12:04:47 +00001369 /* handle inexactly overlapping matches between
1370 * mappings */
1371 if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1372 !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1373 /* new mapping is not a subset of the region */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001374 if (!(capabilities & NOMMU_MAP_DIRECT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375 goto sharing_violation;
1376 continue;
1377 }
1378
David Howells8feae132009-01-08 12:04:47 +00001379 /* we've found a region we can share */
David Howells1e2ae592010-01-15 17:01:33 -08001380 pregion->vm_usage++;
David Howells8feae132009-01-08 12:04:47 +00001381 vma->vm_region = pregion;
1382 start = pregion->vm_start;
1383 start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1384 vma->vm_start = start;
1385 vma->vm_end = start + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386
David Howells8feae132009-01-08 12:04:47 +00001387 if (pregion->vm_flags & VM_MAPPED_COPY) {
1388 kdebug("share copy");
1389 vma->vm_flags |= VM_MAPPED_COPY;
1390 } else {
1391 kdebug("share mmap");
1392 ret = do_mmap_shared_file(vma);
1393 if (ret < 0) {
1394 vma->vm_region = NULL;
1395 vma->vm_start = 0;
1396 vma->vm_end = 0;
David Howells1e2ae592010-01-15 17:01:33 -08001397 pregion->vm_usage--;
David Howells8feae132009-01-08 12:04:47 +00001398 pregion = NULL;
1399 goto error_just_free;
1400 }
1401 }
1402 fput(region->vm_file);
1403 kmem_cache_free(vm_region_jar, region);
1404 region = pregion;
1405 result = start;
1406 goto share;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407 }
1408
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 /* obtain the address at which to make a shared mapping
1410 * - this is the hook for quasi-memory character devices to
1411 * tell us the location of a shared mapping
1412 */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001413 if (capabilities & NOMMU_MAP_DIRECT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 addr = file->f_op->get_unmapped_area(file, addr, len,
1415 pgoff, flags);
Namhyung Kimbb005a52011-05-24 17:11:27 -07001416 if (IS_ERR_VALUE(addr)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 ret = addr;
Namhyung Kimbb005a52011-05-24 17:11:27 -07001418 if (ret != -ENOSYS)
David Howells8feae132009-01-08 12:04:47 +00001419 goto error_just_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420
1421 /* the driver refused to tell us where to site
1422 * the mapping so we'll have to attempt to copy
1423 * it */
Namhyung Kimbb005a52011-05-24 17:11:27 -07001424 ret = -ENODEV;
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001425 if (!(capabilities & NOMMU_MAP_COPY))
David Howells8feae132009-01-08 12:04:47 +00001426 goto error_just_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001428 capabilities &= ~NOMMU_MAP_DIRECT;
David Howells8feae132009-01-08 12:04:47 +00001429 } else {
1430 vma->vm_start = region->vm_start = addr;
1431 vma->vm_end = region->vm_end = addr + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432 }
1433 }
1434 }
1435
David Howells8feae132009-01-08 12:04:47 +00001436 vma->vm_region = region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437
David Howells645d83c2009-09-24 15:13:10 +01001438 /* set up the mapping
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001439 * - the region is filled in if NOMMU_MAP_DIRECT is still set
David Howells645d83c2009-09-24 15:13:10 +01001440 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 if (file && vma->vm_flags & VM_SHARED)
David Howells8feae132009-01-08 12:04:47 +00001442 ret = do_mmap_shared_file(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 else
David Howells645d83c2009-09-24 15:13:10 +01001444 ret = do_mmap_private(vma, region, len, capabilities);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445 if (ret < 0)
David Howells645d83c2009-09-24 15:13:10 +01001446 goto error_just_free;
1447 add_nommu_region(region);
David Howells8feae132009-01-08 12:04:47 +00001448
Jie Zhangea637632009-12-14 18:00:02 -08001449 /* clear anonymous mappings that don't ask for uninitialized data */
1450 if (!vma->vm_file && !(flags & MAP_UNINITIALIZED))
1451 memset((void *)region->vm_start, 0,
1452 region->vm_end - region->vm_start);
1453
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 /* okay... we have a mapping; now we have to register it */
David Howells8feae132009-01-08 12:04:47 +00001455 result = vma->vm_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 current->mm->total_vm += len >> PAGE_SHIFT;
1458
David Howells8feae132009-01-08 12:04:47 +00001459share:
1460 add_vma_to_mm(current->mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461
Mike Frysingercfe79c02010-01-06 17:23:23 +00001462 /* we flush the region from the icache only when the first executable
1463 * mapping of it is made */
1464 if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1465 flush_icache_range(region->vm_start, region->vm_end);
1466 region->vm_icache_flushed = true;
1467 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468
Mike Frysingercfe79c02010-01-06 17:23:23 +00001469 up_write(&nommu_region_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470
David Howells8feae132009-01-08 12:04:47 +00001471 kleave(" = %lx", result);
1472 return result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473
David Howells8feae132009-01-08 12:04:47 +00001474error_just_free:
1475 up_write(&nommu_region_sem);
1476error:
David Howells89a86402009-10-30 13:13:26 +00001477 if (region->vm_file)
1478 fput(region->vm_file);
David Howells8feae132009-01-08 12:04:47 +00001479 kmem_cache_free(vm_region_jar, region);
David Howells89a86402009-10-30 13:13:26 +00001480 if (vma->vm_file)
1481 fput(vma->vm_file);
David Howells8feae132009-01-08 12:04:47 +00001482 kmem_cache_free(vm_area_cachep, vma);
1483 kleave(" = %d", ret);
1484 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485
David Howells8feae132009-01-08 12:04:47 +00001486sharing_violation:
1487 up_write(&nommu_region_sem);
1488 printk(KERN_WARNING "Attempt to share mismatched mappings\n");
1489 ret = -EINVAL;
1490 goto error;
1491
1492error_getting_vma:
1493 kmem_cache_free(vm_region_jar, region);
1494 printk(KERN_WARNING "Allocation of vma for %lu byte allocation"
1495 " from process %d failed\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496 len, current->pid);
David Rientjes7bf02ea2011-05-24 17:11:16 -07001497 show_free_areas(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498 return -ENOMEM;
1499
David Howells8feae132009-01-08 12:04:47 +00001500error_getting_region:
1501 printk(KERN_WARNING "Allocation of vm region for %lu byte allocation"
1502 " from process %d failed\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 len, current->pid);
David Rientjes7bf02ea2011-05-24 17:11:16 -07001504 show_free_areas(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505 return -ENOMEM;
1506}
Linus Torvalds6be5ceb2012-04-20 17:13:58 -07001507
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001508SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1509 unsigned long, prot, unsigned long, flags,
1510 unsigned long, fd, unsigned long, pgoff)
1511{
1512 struct file *file = NULL;
1513 unsigned long retval = -EBADF;
1514
Al Viro120a7952010-10-30 02:54:44 -04001515 audit_mmap_fd(fd, flags);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001516 if (!(flags & MAP_ANONYMOUS)) {
1517 file = fget(fd);
1518 if (!file)
1519 goto out;
1520 }
1521
1522 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1523
Greg Ungererad1ed292012-06-04 14:29:59 +10001524 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001525
1526 if (file)
1527 fput(file);
1528out:
1529 return retval;
1530}
1531
Christoph Hellwiga4679372010-03-10 15:21:15 -08001532#ifdef __ARCH_WANT_SYS_OLD_MMAP
1533struct mmap_arg_struct {
1534 unsigned long addr;
1535 unsigned long len;
1536 unsigned long prot;
1537 unsigned long flags;
1538 unsigned long fd;
1539 unsigned long offset;
1540};
1541
1542SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1543{
1544 struct mmap_arg_struct a;
1545
1546 if (copy_from_user(&a, arg, sizeof(a)))
1547 return -EFAULT;
1548 if (a.offset & ~PAGE_MASK)
1549 return -EINVAL;
1550
1551 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1552 a.offset >> PAGE_SHIFT);
1553}
1554#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1555
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556/*
David Howells8feae132009-01-08 12:04:47 +00001557 * split a vma into two pieces at address 'addr', a new vma is allocated either
1558 * for the first part or the tail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559 */
David Howells8feae132009-01-08 12:04:47 +00001560int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1561 unsigned long addr, int new_below)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562{
David Howells8feae132009-01-08 12:04:47 +00001563 struct vm_area_struct *new;
1564 struct vm_region *region;
1565 unsigned long npages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566
David Howells8feae132009-01-08 12:04:47 +00001567 kenter("");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568
David Howells779c1022010-01-15 17:01:34 -08001569 /* we're only permitted to split anonymous regions (these should have
1570 * only a single usage on the region) */
1571 if (vma->vm_file)
David Howells8feae132009-01-08 12:04:47 +00001572 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573
David Howells8feae132009-01-08 12:04:47 +00001574 if (mm->map_count >= sysctl_max_map_count)
1575 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576
David Howells8feae132009-01-08 12:04:47 +00001577 region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1578 if (!region)
1579 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580
David Howells8feae132009-01-08 12:04:47 +00001581 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1582 if (!new) {
1583 kmem_cache_free(vm_region_jar, region);
1584 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585 }
David Howells8feae132009-01-08 12:04:47 +00001586
1587 /* most fields are the same, copy all, and then fixup */
1588 *new = *vma;
1589 *region = *vma->vm_region;
1590 new->vm_region = region;
1591
1592 npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1593
1594 if (new_below) {
Paul Mundtdd8632a2009-01-08 12:04:47 +00001595 region->vm_top = region->vm_end = new->vm_end = addr;
David Howells8feae132009-01-08 12:04:47 +00001596 } else {
1597 region->vm_start = new->vm_start = addr;
1598 region->vm_pgoff = new->vm_pgoff += npages;
1599 }
1600
1601 if (new->vm_ops && new->vm_ops->open)
1602 new->vm_ops->open(new);
1603
1604 delete_vma_from_mm(vma);
1605 down_write(&nommu_region_sem);
1606 delete_nommu_region(vma->vm_region);
1607 if (new_below) {
1608 vma->vm_region->vm_start = vma->vm_start = addr;
1609 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1610 } else {
1611 vma->vm_region->vm_end = vma->vm_end = addr;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001612 vma->vm_region->vm_top = addr;
David Howells8feae132009-01-08 12:04:47 +00001613 }
1614 add_nommu_region(vma->vm_region);
1615 add_nommu_region(new->vm_region);
1616 up_write(&nommu_region_sem);
1617 add_vma_to_mm(mm, vma);
1618 add_vma_to_mm(mm, new);
1619 return 0;
1620}
1621
1622/*
1623 * shrink a VMA by removing the specified chunk from either the beginning or
1624 * the end
1625 */
1626static int shrink_vma(struct mm_struct *mm,
1627 struct vm_area_struct *vma,
1628 unsigned long from, unsigned long to)
1629{
1630 struct vm_region *region;
1631
1632 kenter("");
1633
1634 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1635 * and list */
1636 delete_vma_from_mm(vma);
1637 if (from > vma->vm_start)
1638 vma->vm_end = from;
1639 else
1640 vma->vm_start = to;
1641 add_vma_to_mm(mm, vma);
1642
1643 /* cut the backing region down to size */
1644 region = vma->vm_region;
David Howells1e2ae592010-01-15 17:01:33 -08001645 BUG_ON(region->vm_usage != 1);
David Howells8feae132009-01-08 12:04:47 +00001646
1647 down_write(&nommu_region_sem);
1648 delete_nommu_region(region);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001649 if (from > region->vm_start) {
1650 to = region->vm_top;
1651 region->vm_top = region->vm_end = from;
1652 } else {
David Howells8feae132009-01-08 12:04:47 +00001653 region->vm_start = to;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001654 }
David Howells8feae132009-01-08 12:04:47 +00001655 add_nommu_region(region);
1656 up_write(&nommu_region_sem);
1657
1658 free_page_series(from, to);
1659 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660}
1661
David Howells30340972006-09-27 01:50:20 -07001662/*
1663 * release a mapping
David Howells8feae132009-01-08 12:04:47 +00001664 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1665 * VMA, though it need not cover the whole VMA
David Howells30340972006-09-27 01:50:20 -07001666 */
David Howells8feae132009-01-08 12:04:47 +00001667int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668{
David Howells8feae132009-01-08 12:04:47 +00001669 struct vm_area_struct *vma;
Bob Liuf67d9b12011-05-24 17:12:56 -07001670 unsigned long end;
David Howells8feae132009-01-08 12:04:47 +00001671 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672
David Howells8feae132009-01-08 12:04:47 +00001673 kenter(",%lx,%zx", start, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674
Bob Liuf67d9b12011-05-24 17:12:56 -07001675 len = PAGE_ALIGN(len);
David Howells8feae132009-01-08 12:04:47 +00001676 if (len == 0)
1677 return -EINVAL;
1678
Bob Liuf67d9b12011-05-24 17:12:56 -07001679 end = start + len;
1680
David Howells8feae132009-01-08 12:04:47 +00001681 /* find the first potentially overlapping VMA */
1682 vma = find_vma(mm, start);
1683 if (!vma) {
Choi Gi-yongac714902014-04-07 15:37:36 -07001684 static int limit;
David Howells33e5d7692009-04-02 16:56:32 -07001685 if (limit < 5) {
1686 printk(KERN_WARNING
1687 "munmap of memory not mmapped by process %d"
1688 " (%s): 0x%lx-0x%lx\n",
1689 current->pid, current->comm,
1690 start, start + len - 1);
1691 limit++;
1692 }
David Howells8feae132009-01-08 12:04:47 +00001693 return -EINVAL;
David Howells30340972006-09-27 01:50:20 -07001694 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695
David Howells8feae132009-01-08 12:04:47 +00001696 /* we're allowed to split an anonymous VMA but not a file-backed one */
1697 if (vma->vm_file) {
1698 do {
1699 if (start > vma->vm_start) {
1700 kleave(" = -EINVAL [miss]");
1701 return -EINVAL;
1702 }
1703 if (end == vma->vm_end)
1704 goto erase_whole_vma;
Namhyung Kimd75a3102011-05-24 17:11:25 -07001705 vma = vma->vm_next;
1706 } while (vma);
David Howells8feae132009-01-08 12:04:47 +00001707 kleave(" = -EINVAL [split file]");
1708 return -EINVAL;
1709 } else {
1710 /* the chunk must be a subset of the VMA found */
1711 if (start == vma->vm_start && end == vma->vm_end)
1712 goto erase_whole_vma;
1713 if (start < vma->vm_start || end > vma->vm_end) {
1714 kleave(" = -EINVAL [superset]");
1715 return -EINVAL;
1716 }
1717 if (start & ~PAGE_MASK) {
1718 kleave(" = -EINVAL [unaligned start]");
1719 return -EINVAL;
1720 }
1721 if (end != vma->vm_end && end & ~PAGE_MASK) {
1722 kleave(" = -EINVAL [unaligned split]");
1723 return -EINVAL;
1724 }
1725 if (start != vma->vm_start && end != vma->vm_end) {
1726 ret = split_vma(mm, vma, start, 1);
1727 if (ret < 0) {
1728 kleave(" = %d [split]", ret);
1729 return ret;
1730 }
1731 }
1732 return shrink_vma(mm, vma, start, end);
1733 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734
David Howells8feae132009-01-08 12:04:47 +00001735erase_whole_vma:
1736 delete_vma_from_mm(vma);
1737 delete_vma(mm, vma);
1738 kleave(" = 0");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739 return 0;
1740}
Paul Mundtb5073172007-07-21 04:37:25 -07001741EXPORT_SYMBOL(do_munmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742
Al Virobfce2812012-04-20 21:57:04 -04001743int vm_munmap(unsigned long addr, size_t len)
David Howells30340972006-09-27 01:50:20 -07001744{
Al Virobfce2812012-04-20 21:57:04 -04001745 struct mm_struct *mm = current->mm;
David Howells30340972006-09-27 01:50:20 -07001746 int ret;
David Howells30340972006-09-27 01:50:20 -07001747
1748 down_write(&mm->mmap_sem);
1749 ret = do_munmap(mm, addr, len);
1750 up_write(&mm->mmap_sem);
1751 return ret;
1752}
Linus Torvaldsa46ef992012-04-20 16:20:01 -07001753EXPORT_SYMBOL(vm_munmap);
1754
1755SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1756{
Al Virobfce2812012-04-20 21:57:04 -04001757 return vm_munmap(addr, len);
Linus Torvaldsa46ef992012-04-20 16:20:01 -07001758}
David Howells30340972006-09-27 01:50:20 -07001759
1760/*
David Howells8feae132009-01-08 12:04:47 +00001761 * release all the mappings made in a process's VM space
David Howells30340972006-09-27 01:50:20 -07001762 */
David Howells8feae132009-01-08 12:04:47 +00001763void exit_mmap(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764{
David Howells8feae132009-01-08 12:04:47 +00001765 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766
David Howells8feae132009-01-08 12:04:47 +00001767 if (!mm)
1768 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769
David Howells8feae132009-01-08 12:04:47 +00001770 kenter("");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771
David Howells8feae132009-01-08 12:04:47 +00001772 mm->total_vm = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773
David Howells8feae132009-01-08 12:04:47 +00001774 while ((vma = mm->mmap)) {
1775 mm->mmap = vma->vm_next;
1776 delete_vma_from_mm(vma);
1777 delete_vma(mm, vma);
Steven J. Magnani04c34962010-11-24 12:56:54 -08001778 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779 }
David Howells8feae132009-01-08 12:04:47 +00001780
1781 kleave("");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782}
1783
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07001784unsigned long vm_brk(unsigned long addr, unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785{
1786 return -ENOMEM;
1787}
1788
1789/*
David Howells6fa5f802006-09-27 01:50:21 -07001790 * expand (or shrink) an existing mapping, potentially moving it at the same
1791 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792 *
David Howells6fa5f802006-09-27 01:50:21 -07001793 * under NOMMU conditions, we only permit changing a mapping's size, and only
David Howells8feae132009-01-08 12:04:47 +00001794 * as long as it stays within the region allocated by do_mmap_private() and the
1795 * block is not shareable
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796 *
David Howells6fa5f802006-09-27 01:50:21 -07001797 * MREMAP_FIXED is not supported under NOMMU conditions
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798 */
Al Viro4b377ba2013-03-04 10:47:59 -05001799static unsigned long do_mremap(unsigned long addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800 unsigned long old_len, unsigned long new_len,
1801 unsigned long flags, unsigned long new_addr)
1802{
David Howells6fa5f802006-09-27 01:50:21 -07001803 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804
1805 /* insanity checks first */
Bob Liuf67d9b12011-05-24 17:12:56 -07001806 old_len = PAGE_ALIGN(old_len);
1807 new_len = PAGE_ALIGN(new_len);
David Howells8feae132009-01-08 12:04:47 +00001808 if (old_len == 0 || new_len == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809 return (unsigned long) -EINVAL;
1810
David Howells8feae132009-01-08 12:04:47 +00001811 if (addr & ~PAGE_MASK)
1812 return -EINVAL;
1813
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814 if (flags & MREMAP_FIXED && new_addr != addr)
1815 return (unsigned long) -EINVAL;
1816
David Howells8feae132009-01-08 12:04:47 +00001817 vma = find_vma_exact(current->mm, addr, old_len);
David Howells6fa5f802006-09-27 01:50:21 -07001818 if (!vma)
1819 return (unsigned long) -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820
David Howells6fa5f802006-09-27 01:50:21 -07001821 if (vma->vm_end != vma->vm_start + old_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822 return (unsigned long) -EFAULT;
1823
David Howells6fa5f802006-09-27 01:50:21 -07001824 if (vma->vm_flags & VM_MAYSHARE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825 return (unsigned long) -EPERM;
1826
David Howells8feae132009-01-08 12:04:47 +00001827 if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828 return (unsigned long) -ENOMEM;
1829
1830 /* all checks complete - do it */
David Howells6fa5f802006-09-27 01:50:21 -07001831 vma->vm_end = vma->vm_start + new_len;
David Howells6fa5f802006-09-27 01:50:21 -07001832 return vma->vm_start;
1833}
1834
Heiko Carstens6a6160a2009-01-14 14:14:15 +01001835SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1836 unsigned long, new_len, unsigned long, flags,
1837 unsigned long, new_addr)
David Howells6fa5f802006-09-27 01:50:21 -07001838{
1839 unsigned long ret;
1840
1841 down_write(&current->mm->mmap_sem);
1842 ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1843 up_write(&current->mm->mmap_sem);
1844 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845}
1846
Michel Lespinasse240aade2013-02-22 16:35:56 -08001847struct page *follow_page_mask(struct vm_area_struct *vma,
1848 unsigned long address, unsigned int flags,
1849 unsigned int *page_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850{
Michel Lespinasse240aade2013-02-22 16:35:56 -08001851 *page_mask = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852 return NULL;
1853}
1854
Bob Liu8f3b1322011-07-08 15:39:46 -07001855int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1856 unsigned long pfn, unsigned long size, pgprot_t prot)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857{
Bob Liu8f3b1322011-07-08 15:39:46 -07001858 if (addr != (pfn << PAGE_SHIFT))
1859 return -EINVAL;
1860
Konstantin Khlebnikov314e51b2012-10-08 16:29:02 -07001861 vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
Greg Ungerer66aa2b42005-09-12 11:18:10 +10001862 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863}
Luke Yang22c4af42006-07-14 00:24:09 -07001864EXPORT_SYMBOL(remap_pfn_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865
Linus Torvalds3c0b9de2013-04-27 13:25:38 -07001866int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)
1867{
1868 unsigned long pfn = start >> PAGE_SHIFT;
1869 unsigned long vm_len = vma->vm_end - vma->vm_start;
1870
1871 pfn += vma->vm_pgoff;
1872 return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);
1873}
1874EXPORT_SYMBOL(vm_iomap_memory);
1875
Paul Mundtf905bc42008-02-04 22:29:59 -08001876int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1877 unsigned long pgoff)
1878{
1879 unsigned int size = vma->vm_end - vma->vm_start;
1880
1881 if (!(vma->vm_flags & VM_USERMAP))
1882 return -EINVAL;
1883
1884 vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1885 vma->vm_end = vma->vm_start + size;
1886
1887 return 0;
1888}
1889EXPORT_SYMBOL(remap_vmalloc_range);
1890
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1892 unsigned long len, unsigned long pgoff, unsigned long flags)
1893{
1894 return -ENOMEM;
1895}
1896
Linus Torvalds1da177e2005-04-16 15:20:36 -07001897void unmap_mapping_range(struct address_space *mapping,
1898 loff_t const holebegin, loff_t const holelen,
1899 int even_cows)
1900{
1901}
Luke Yang22c4af42006-07-14 00:24:09 -07001902EXPORT_SYMBOL(unmap_mapping_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903
1904/*
1905 * Check that a process has enough memory to allocate a new virtual
1906 * mapping. 0 means there is enough memory for the allocation to
1907 * succeed and -ENOMEM implies there is not.
1908 *
1909 * We currently support three overcommit policies, which are set via the
1910 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
1911 *
1912 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1913 * Additional code 2002 Jul 20 by Robert Love.
1914 *
1915 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1916 *
1917 * Note this is a helper function intended to be used by LSMs which
1918 * wish to use this logic.
1919 */
Alan Cox34b4e4a2007-08-22 14:01:28 -07001920int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921{
Roman Gushchin8138a672015-02-11 15:28:42 -08001922 long free, allowed, reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923
1924 vm_acct_memory(pages);
1925
1926 /*
1927 * Sometimes we want to use more memory than we have
1928 */
1929 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1930 return 0;
1931
1932 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
Dmitry Finkc15bef32011-07-25 17:12:19 -07001933 free = global_page_state(NR_FREE_PAGES);
1934 free += global_page_state(NR_FILE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935
Dmitry Finkc15bef32011-07-25 17:12:19 -07001936 /*
1937 * shmem pages shouldn't be counted as free in this
1938 * case, they can't be purged, only swapped out, and
1939 * that won't affect the overall amount of available
1940 * memory in the system.
1941 */
1942 free -= global_page_state(NR_SHMEM);
1943
Shaohua Liec8acf22013-02-22 16:34:38 -08001944 free += get_nr_swap_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945
1946 /*
1947 * Any slabs which are created with the
1948 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1949 * which are reclaimable, under pressure. The dentry
1950 * cache and most inode caches should fall into this
1951 */
Christoph Lameter972d1a72006-09-25 23:31:51 -07001952 free += global_page_state(NR_SLAB_RECLAIMABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953
1954 /*
Dmitry Finkc15bef32011-07-25 17:12:19 -07001955 * Leave reserved pages. The pages are not for anonymous pages.
1956 */
1957 if (free <= totalreserve_pages)
1958 goto error;
1959 else
1960 free -= totalreserve_pages;
1961
1962 /*
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07001963 * Reserve some for root
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964 */
1965 if (!cap_sys_admin)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07001966 free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967
1968 if (free > pages)
1969 return 0;
1970
Hideo AOKId5ddc792006-04-10 22:53:01 -07001971 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972 }
1973
Jerome Marchand00619bc2013-11-12 15:08:31 -08001974 allowed = vm_commit_limit();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975 /*
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07001976 * Reserve some 3% for root
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977 */
1978 if (!cap_sys_admin)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07001979 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07001981 /*
1982 * Don't let a single process grow so big a user can't recover
1983 */
1984 if (mm) {
1985 reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
Roman Gushchin8138a672015-02-11 15:28:42 -08001986 allowed -= min_t(long, mm->total_vm / 32, reserve);
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07001987 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07001989 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990 return 0;
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07001991
Hideo AOKId5ddc792006-04-10 22:53:01 -07001992error:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993 vm_unacct_memory(pages);
1994
1995 return -ENOMEM;
1996}
1997
Nick Piggind0217ac2007-07-19 01:47:03 -07001998int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
David Howellsb0e15192006-01-06 00:11:42 -08001999{
2000 BUG();
Nick Piggind0217ac2007-07-19 01:47:03 -07002001 return 0;
David Howellsb0e15192006-01-06 00:11:42 -08002002}
Paul Mundtb5073172007-07-21 04:37:25 -07002003EXPORT_SYMBOL(filemap_fault);
David Howells0ec76a12006-09-27 01:50:15 -07002004
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002005void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf)
2006{
2007 BUG();
2008}
2009EXPORT_SYMBOL(filemap_map_pages);
2010
Mike Frysingerf55f1992011-03-29 14:05:12 +01002011static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
2012 unsigned long addr, void *buf, int len, int write)
David Howells0ec76a12006-09-27 01:50:15 -07002013{
David Howells0ec76a12006-09-27 01:50:15 -07002014 struct vm_area_struct *vma;
David Howells0ec76a12006-09-27 01:50:15 -07002015
2016 down_read(&mm->mmap_sem);
2017
2018 /* the access must start within one of the target process's mappings */
David Howells0159b142006-09-27 01:50:16 -07002019 vma = find_vma(mm, addr);
2020 if (vma) {
David Howells0ec76a12006-09-27 01:50:15 -07002021 /* don't overrun this mapping */
2022 if (addr + len >= vma->vm_end)
2023 len = vma->vm_end - addr;
2024
2025 /* only read or write mappings where it is permitted */
David Howellsd00c7b992006-09-27 01:50:19 -07002026 if (write && vma->vm_flags & VM_MAYWRITE)
Jie Zhang79597222010-01-06 17:23:28 +00002027 copy_to_user_page(vma, NULL, addr,
2028 (void *) addr, buf, len);
David Howellsd00c7b992006-09-27 01:50:19 -07002029 else if (!write && vma->vm_flags & VM_MAYREAD)
Jie Zhang79597222010-01-06 17:23:28 +00002030 copy_from_user_page(vma, NULL, addr,
2031 buf, (void *) addr, len);
David Howells0ec76a12006-09-27 01:50:15 -07002032 else
2033 len = 0;
2034 } else {
2035 len = 0;
2036 }
2037
2038 up_read(&mm->mmap_sem);
Mike Frysingerf55f1992011-03-29 14:05:12 +01002039
2040 return len;
2041}
2042
2043/**
2044 * @access_remote_vm - access another process' address space
2045 * @mm: the mm_struct of the target address space
2046 * @addr: start address to access
2047 * @buf: source or destination buffer
2048 * @len: number of bytes to transfer
2049 * @write: whether the access is a write
2050 *
2051 * The caller must hold a reference on @mm.
2052 */
2053int access_remote_vm(struct mm_struct *mm, unsigned long addr,
2054 void *buf, int len, int write)
2055{
2056 return __access_remote_vm(NULL, mm, addr, buf, len, write);
2057}
2058
2059/*
2060 * Access another process' address space.
2061 * - source/target buffer must be kernel space
2062 */
2063int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
2064{
2065 struct mm_struct *mm;
2066
2067 if (addr + len < addr)
2068 return 0;
2069
2070 mm = get_task_mm(tsk);
2071 if (!mm)
2072 return 0;
2073
2074 len = __access_remote_vm(tsk, mm, addr, buf, len, write);
2075
David Howells0ec76a12006-09-27 01:50:15 -07002076 mmput(mm);
2077 return len;
2078}
David Howells7e660872010-01-15 17:01:39 -08002079
2080/**
2081 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
2082 * @inode: The inode to check
2083 * @size: The current filesize of the inode
2084 * @newsize: The proposed filesize of the inode
2085 *
2086 * Check the shared mappings on an inode on behalf of a shrinking truncate to
2087 * make sure that that any outstanding VMAs aren't broken and then shrink the
2088 * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
2089 * automatically grant mappings that are too large.
2090 */
2091int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
2092 size_t newsize)
2093{
2094 struct vm_area_struct *vma;
David Howells7e660872010-01-15 17:01:39 -08002095 struct vm_region *region;
2096 pgoff_t low, high;
2097 size_t r_size, r_top;
2098
2099 low = newsize >> PAGE_SHIFT;
2100 high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
2101
2102 down_write(&nommu_region_sem);
Davidlohr Bueso1acf2e02014-12-12 16:54:39 -08002103 i_mmap_lock_read(inode->i_mapping);
David Howells7e660872010-01-15 17:01:39 -08002104
2105 /* search for VMAs that fall within the dead zone */
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -07002106 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) {
David Howells7e660872010-01-15 17:01:39 -08002107 /* found one - only interested if it's shared out of the page
2108 * cache */
2109 if (vma->vm_flags & VM_SHARED) {
Davidlohr Bueso1acf2e02014-12-12 16:54:39 -08002110 i_mmap_unlock_read(inode->i_mapping);
David Howells7e660872010-01-15 17:01:39 -08002111 up_write(&nommu_region_sem);
2112 return -ETXTBSY; /* not quite true, but near enough */
2113 }
2114 }
2115
2116 /* reduce any regions that overlap the dead zone - if in existence,
2117 * these will be pointed to by VMAs that don't overlap the dead zone
2118 *
2119 * we don't check for any regions that start beyond the EOF as there
2120 * shouldn't be any
2121 */
Davidlohr Bueso1acf2e02014-12-12 16:54:39 -08002122 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, 0, ULONG_MAX) {
David Howells7e660872010-01-15 17:01:39 -08002123 if (!(vma->vm_flags & VM_SHARED))
2124 continue;
2125
2126 region = vma->vm_region;
2127 r_size = region->vm_top - region->vm_start;
2128 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
2129
2130 if (r_top > newsize) {
2131 region->vm_top -= r_top - newsize;
2132 if (region->vm_end > region->vm_top)
2133 region->vm_end = region->vm_top;
2134 }
2135 }
2136
Davidlohr Bueso1acf2e02014-12-12 16:54:39 -08002137 i_mmap_unlock_read(inode->i_mapping);
David Howells7e660872010-01-15 17:01:39 -08002138 up_write(&nommu_region_sem);
2139 return 0;
2140}
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07002141
2142/*
2143 * Initialise sysctl_user_reserve_kbytes.
2144 *
2145 * This is intended to prevent a user from starting a single memory hogging
2146 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
2147 * mode.
2148 *
2149 * The default value is min(3% of free memory, 128MB)
2150 * 128MB is enough to recover with sshd/login, bash, and top/kill.
2151 */
2152static int __meminit init_user_reserve(void)
2153{
2154 unsigned long free_kbytes;
2155
2156 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
2157
2158 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
2159 return 0;
2160}
2161module_init(init_user_reserve)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07002162
2163/*
2164 * Initialise sysctl_admin_reserve_kbytes.
2165 *
2166 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
2167 * to log in and kill a memory hogging process.
2168 *
2169 * Systems with more than 256MB will reserve 8MB, enough to recover
2170 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
2171 * only reserve 3% of free pages by default.
2172 */
2173static int __meminit init_admin_reserve(void)
2174{
2175 unsigned long free_kbytes;
2176
2177 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
2178
2179 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
2180 return 0;
2181}
2182module_init(init_admin_reserve)