blob: e544508e2a4bc3e3dfa6190abdd3faf0616c95f2 [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;
gchen gchen5b8bf302015-03-12 16:26:05 -070065EXPORT_SYMBOL(max_mapnr);
Hugh Dickins4266c972009-09-23 17:05:53 +010066unsigned long highest_memmap_pfn;
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -070067struct percpu_counter vm_committed_as;
Linus Torvalds1da177e2005-04-16 15:20:36 -070068int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
69int sysctl_overcommit_ratio = 50; /* default is 50% */
Jerome Marchand49f0ce52014-01-21 15:49:14 -080070unsigned long sysctl_overcommit_kbytes __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070071int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
David Howellsfc4d5c22009-05-06 16:03:05 -070072int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -070073unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -070074unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
Linus Torvalds1da177e2005-04-16 15:20:36 -070075int heap_stack_gap = 0;
76
David Howells33e5d7692009-04-02 16:56:32 -070077atomic_long_t mmap_pages_allocated;
David Howells8feae132009-01-08 12:04:47 +000078
K. Y. Srinivasan997071b2012-11-15 14:34:42 -080079/*
80 * The global memory commitment made in the system can be a metric
81 * that can be used to drive ballooning decisions when Linux is hosted
82 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
83 * balancing memory across competing virtual machines that are hosted.
84 * Several metrics drive this policy engine including the guest reported
85 * memory commitment.
86 */
87unsigned long vm_memory_committed(void)
88{
89 return percpu_counter_read_positive(&vm_committed_as);
90}
91
92EXPORT_SYMBOL_GPL(vm_memory_committed);
93
Linus Torvalds1da177e2005-04-16 15:20:36 -070094EXPORT_SYMBOL(mem_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -070095
David Howells8feae132009-01-08 12:04:47 +000096/* list of mapped, potentially shareable regions */
97static struct kmem_cache *vm_region_jar;
98struct rb_root nommu_region_tree = RB_ROOT;
99DECLARE_RWSEM(nommu_region_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +0400101const struct vm_operations_struct generic_file_vm_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102};
103
104/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105 * Return the total memory allocated for this pointer, not
106 * just what the caller asked for.
107 *
108 * Doesn't have to be accurate, i.e. may have races.
109 */
110unsigned int kobjsize(const void *objp)
111{
112 struct page *page;
113
Michael Hennerich4016a132008-04-28 02:13:38 -0700114 /*
115 * If the object we have should not have ksize performed on it,
116 * return size of 0
117 */
Paul Mundt5a1603b2008-06-12 16:29:55 +0900118 if (!objp || !virt_addr_valid(objp))
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700119 return 0;
120
121 page = virt_to_head_page(objp);
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700122
123 /*
124 * If the allocator sets PageSlab, we know the pointer came from
125 * kmalloc().
126 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127 if (PageSlab(page))
128 return ksize(objp);
129
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700130 /*
Paul Mundtab2e83e2009-01-08 12:04:48 +0000131 * If it's not a compound page, see if we have a matching VMA
132 * region. This test is intentionally done in reverse order,
133 * so if there's no VMA, we still fall through and hand back
134 * PAGE_SIZE for 0-order pages.
135 */
136 if (!PageCompound(page)) {
137 struct vm_area_struct *vma;
138
139 vma = find_vma(current->mm, (unsigned long)objp);
140 if (vma)
141 return vma->vm_end - vma->vm_start;
142 }
143
144 /*
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700145 * The ksize() function is only guaranteed to work for pointers
Paul Mundt5a1603b2008-06-12 16:29:55 +0900146 * returned by kmalloc(). So handle arbitrary pointers here.
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700147 */
Paul Mundt5a1603b2008-06-12 16:29:55 +0900148 return PAGE_SIZE << compound_order(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149}
150
Michel Lespinasse28a35712013-02-22 16:35:55 -0800151long __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
152 unsigned long start, unsigned long nr_pages,
153 unsigned int foll_flags, struct page **pages,
154 struct vm_area_struct **vmas, int *nonblocking)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155{
Sonic Zhang910e46d2006-09-27 01:50:17 -0700156 struct vm_area_struct *vma;
David Howells7b4d5b82006-09-27 01:50:18 -0700157 unsigned long vm_flags;
158 int i;
159
160 /* calculate required read or write permissions.
Hugh Dickins58fa8792009-09-21 17:03:31 -0700161 * If FOLL_FORCE is set, we only require the "MAY" flags.
David Howells7b4d5b82006-09-27 01:50:18 -0700162 */
Hugh Dickins58fa8792009-09-21 17:03:31 -0700163 vm_flags = (foll_flags & FOLL_WRITE) ?
164 (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
165 vm_flags &= (foll_flags & FOLL_FORCE) ?
166 (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167
Peter Zijlstra9d737772009-06-25 11:58:55 +0200168 for (i = 0; i < nr_pages; i++) {
David Howells7561e8c2010-03-25 16:48:38 +0000169 vma = find_vma(mm, start);
David Howells7b4d5b82006-09-27 01:50:18 -0700170 if (!vma)
171 goto finish_or_fault;
172
173 /* protect what we can, including chardevs */
Hugh Dickins1c3aff12009-09-21 17:03:24 -0700174 if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
175 !(vm_flags & vma->vm_flags))
David Howells7b4d5b82006-09-27 01:50:18 -0700176 goto finish_or_fault;
Sonic Zhang910e46d2006-09-27 01:50:17 -0700177
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 if (pages) {
179 pages[i] = virt_to_page(start);
180 if (pages[i])
181 page_cache_get(pages[i]);
182 }
183 if (vmas)
Sonic Zhang910e46d2006-09-27 01:50:17 -0700184 vmas[i] = vma;
David Howellse1ee65d2010-03-25 16:48:44 +0000185 start = (start + PAGE_SIZE) & PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 }
David Howells7b4d5b82006-09-27 01:50:18 -0700187
188 return i;
189
190finish_or_fault:
191 return i ? : -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192}
Nick Pigginb291f002008-10-18 20:26:44 -0700193
Nick Pigginb291f002008-10-18 20:26:44 -0700194/*
195 * get a list of pages in an address range belonging to the specified process
196 * and indicate the VMA that covers each page
197 * - this is potentially dodgy as we may end incrementing the page count of a
198 * slab page or a secondary page from a compound page
199 * - don't permit access to VMAs that don't support it, such as I/O mappings
200 */
Michel Lespinasse28a35712013-02-22 16:35:55 -0800201long get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
202 unsigned long start, unsigned long nr_pages,
203 int write, int force, struct page **pages,
204 struct vm_area_struct **vmas)
Nick Pigginb291f002008-10-18 20:26:44 -0700205{
206 int flags = 0;
207
208 if (write)
Hugh Dickins58fa8792009-09-21 17:03:31 -0700209 flags |= FOLL_WRITE;
Nick Pigginb291f002008-10-18 20:26:44 -0700210 if (force)
Hugh Dickins58fa8792009-09-21 17:03:31 -0700211 flags |= FOLL_FORCE;
Nick Pigginb291f002008-10-18 20:26:44 -0700212
Michel Lespinasse53a77062011-01-13 15:46:14 -0800213 return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas,
214 NULL);
Nick Pigginb291f002008-10-18 20:26:44 -0700215}
Greg Ungerer66aa2b42005-09-12 11:18:10 +1000216EXPORT_SYMBOL(get_user_pages);
217
Andrea Arcangelif0818f42015-02-11 15:27:17 -0800218long get_user_pages_locked(struct task_struct *tsk, struct mm_struct *mm,
219 unsigned long start, unsigned long nr_pages,
220 int write, int force, struct page **pages,
221 int *locked)
222{
223 return get_user_pages(tsk, mm, start, nr_pages, write, force,
224 pages, NULL);
225}
226EXPORT_SYMBOL(get_user_pages_locked);
227
Andrea Arcangeli0fd71a52015-02-11 15:27:20 -0800228long __get_user_pages_unlocked(struct task_struct *tsk, struct mm_struct *mm,
229 unsigned long start, unsigned long nr_pages,
230 int write, int force, struct page **pages,
231 unsigned int gup_flags)
Andrea Arcangelif0818f42015-02-11 15:27:17 -0800232{
233 long ret;
234 down_read(&mm->mmap_sem);
235 ret = get_user_pages(tsk, mm, start, nr_pages, write, force,
236 pages, NULL);
237 up_read(&mm->mmap_sem);
238 return ret;
239}
Andrea Arcangeli0fd71a52015-02-11 15:27:20 -0800240EXPORT_SYMBOL(__get_user_pages_unlocked);
241
242long get_user_pages_unlocked(struct task_struct *tsk, struct mm_struct *mm,
243 unsigned long start, unsigned long nr_pages,
244 int write, int force, struct page **pages)
245{
246 return __get_user_pages_unlocked(tsk, mm, start, nr_pages, write,
247 force, pages, 0);
248}
Andrea Arcangelif0818f42015-02-11 15:27:17 -0800249EXPORT_SYMBOL(get_user_pages_unlocked);
250
Paul Mundtdfc2f912009-06-26 04:31:57 +0900251/**
252 * follow_pfn - look up PFN at a user virtual address
253 * @vma: memory mapping
254 * @address: user virtual address
255 * @pfn: location to store found PFN
256 *
257 * Only IO mappings and raw PFN mappings are allowed.
258 *
259 * Returns zero and the pfn at @pfn on success, -ve otherwise.
260 */
261int follow_pfn(struct vm_area_struct *vma, unsigned long address,
262 unsigned long *pfn)
263{
264 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
265 return -EINVAL;
266
267 *pfn = address >> PAGE_SHIFT;
268 return 0;
269}
270EXPORT_SYMBOL(follow_pfn);
271
Joonsoo Kimf1c40692013-04-29 15:07:37 -0700272LIST_HEAD(vmap_area_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800274void vfree(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275{
276 kfree(addr);
277}
Paul Mundtb5073172007-07-21 04:37:25 -0700278EXPORT_SYMBOL(vfree);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279
Al Virodd0fc662005-10-07 07:46:04 +0100280void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281{
282 /*
Robert P. J. Day85186092007-10-19 23:11:38 +0200283 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
284 * returns only a logical address.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 */
Nick Piggin84097512006-03-22 00:08:34 -0800286 return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287}
Paul Mundtb5073172007-07-21 04:37:25 -0700288EXPORT_SYMBOL(__vmalloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289
Paul Mundtf905bc42008-02-04 22:29:59 -0800290void *vmalloc_user(unsigned long size)
291{
292 void *ret;
293
294 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
295 PAGE_KERNEL);
296 if (ret) {
297 struct vm_area_struct *vma;
298
299 down_write(&current->mm->mmap_sem);
300 vma = find_vma(current->mm, (unsigned long)ret);
301 if (vma)
302 vma->vm_flags |= VM_USERMAP;
303 up_write(&current->mm->mmap_sem);
304 }
305
306 return ret;
307}
308EXPORT_SYMBOL(vmalloc_user);
309
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800310struct page *vmalloc_to_page(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311{
312 return virt_to_page(addr);
313}
Paul Mundtb5073172007-07-21 04:37:25 -0700314EXPORT_SYMBOL(vmalloc_to_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800316unsigned long vmalloc_to_pfn(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317{
318 return page_to_pfn(virt_to_page(addr));
319}
Paul Mundtb5073172007-07-21 04:37:25 -0700320EXPORT_SYMBOL(vmalloc_to_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321
322long vread(char *buf, char *addr, unsigned long count)
323{
Chen Gang9bde9162013-07-03 15:02:36 -0700324 /* Don't allow overflow */
325 if ((unsigned long) buf + count < count)
326 count = -(unsigned long) buf;
327
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 memcpy(buf, addr, count);
329 return count;
330}
331
332long vwrite(char *buf, char *addr, unsigned long count)
333{
334 /* Don't allow overflow */
335 if ((unsigned long) addr + count < count)
336 count = -(unsigned long) addr;
337
338 memcpy(addr, buf, count);
Choi Gi-yongac714902014-04-07 15:37:36 -0700339 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340}
341
342/*
343 * vmalloc - allocate virtually continguos memory
344 *
345 * @size: allocation size
346 *
347 * Allocate enough pages to cover @size from the page level
348 * allocator and map them into continguos kernel virtual space.
349 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +0200350 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 * use __vmalloc() instead.
352 */
353void *vmalloc(unsigned long size)
354{
355 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
356}
Andrew Mortonf6138882006-02-28 16:59:18 -0800357EXPORT_SYMBOL(vmalloc);
358
Dave Younge1ca7782010-10-26 14:22:06 -0700359/*
360 * vzalloc - allocate virtually continguos memory with zero fill
361 *
362 * @size: allocation size
363 *
364 * Allocate enough pages to cover @size from the page level
365 * allocator and map them into continguos kernel virtual space.
366 * The memory allocated is set to zero.
367 *
368 * For tight control over page level allocator and protection flags
369 * use __vmalloc() instead.
370 */
371void *vzalloc(unsigned long size)
372{
373 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
374 PAGE_KERNEL);
375}
376EXPORT_SYMBOL(vzalloc);
377
378/**
379 * vmalloc_node - allocate memory on a specific node
380 * @size: allocation size
381 * @node: numa node
382 *
383 * Allocate enough pages to cover @size from the page level
384 * allocator and map them into contiguous kernel virtual space.
385 *
386 * For tight control over page level allocator and protection flags
387 * use __vmalloc() instead.
388 */
Andrew Mortonf6138882006-02-28 16:59:18 -0800389void *vmalloc_node(unsigned long size, int node)
390{
391 return vmalloc(size);
392}
Paul Mundt9a14f652010-12-24 11:50:34 +0900393EXPORT_SYMBOL(vmalloc_node);
Dave Younge1ca7782010-10-26 14:22:06 -0700394
395/**
396 * vzalloc_node - allocate memory on a specific node with zero fill
397 * @size: allocation size
398 * @node: numa node
399 *
400 * Allocate enough pages to cover @size from the page level
401 * allocator and map them into contiguous kernel virtual space.
402 * The memory allocated is set to zero.
403 *
404 * For tight control over page level allocator and protection flags
405 * use __vmalloc() instead.
406 */
407void *vzalloc_node(unsigned long size, int node)
408{
409 return vzalloc(size);
410}
411EXPORT_SYMBOL(vzalloc_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412
Paul Mundt1af446e2008-08-04 16:01:47 +0900413#ifndef PAGE_KERNEL_EXEC
414# define PAGE_KERNEL_EXEC PAGE_KERNEL
415#endif
416
417/**
418 * vmalloc_exec - allocate virtually contiguous, executable memory
419 * @size: allocation size
420 *
421 * Kernel-internal function to allocate enough pages to cover @size
422 * the page level allocator and map them into contiguous and
423 * executable kernel virtual space.
424 *
425 * For tight control over page level allocator and protection flags
426 * use __vmalloc() instead.
427 */
428
429void *vmalloc_exec(unsigned long size)
430{
431 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
432}
433
Paul Mundtb5073172007-07-21 04:37:25 -0700434/**
435 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 * @size: allocation size
437 *
438 * Allocate enough 32bit PA addressable pages to cover @size from the
439 * page level allocator and map them into continguos kernel virtual space.
440 */
441void *vmalloc_32(unsigned long size)
442{
443 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
444}
Paul Mundtb5073172007-07-21 04:37:25 -0700445EXPORT_SYMBOL(vmalloc_32);
446
447/**
448 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
449 * @size: allocation size
450 *
451 * The resulting memory area is 32bit addressable and zeroed so it can be
452 * mapped to userspace without leaking data.
Paul Mundtf905bc42008-02-04 22:29:59 -0800453 *
454 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
455 * remap_vmalloc_range() are permissible.
Paul Mundtb5073172007-07-21 04:37:25 -0700456 */
457void *vmalloc_32_user(unsigned long size)
458{
Paul Mundtf905bc42008-02-04 22:29:59 -0800459 /*
460 * We'll have to sort out the ZONE_DMA bits for 64-bit,
461 * but for now this can simply use vmalloc_user() directly.
462 */
463 return vmalloc_user(size);
Paul Mundtb5073172007-07-21 04:37:25 -0700464}
465EXPORT_SYMBOL(vmalloc_32_user);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466
467void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
468{
469 BUG();
470 return NULL;
471}
Paul Mundtb5073172007-07-21 04:37:25 -0700472EXPORT_SYMBOL(vmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800474void vunmap(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475{
476 BUG();
477}
Paul Mundtb5073172007-07-21 04:37:25 -0700478EXPORT_SYMBOL(vunmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479
Paul Mundteb6434d2009-01-21 17:45:47 +0900480void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
481{
482 BUG();
483 return NULL;
484}
485EXPORT_SYMBOL(vm_map_ram);
486
487void vm_unmap_ram(const void *mem, unsigned int count)
488{
489 BUG();
490}
491EXPORT_SYMBOL(vm_unmap_ram);
492
493void vm_unmap_aliases(void)
494{
495}
496EXPORT_SYMBOL_GPL(vm_unmap_aliases);
497
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498/*
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -0700499 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
500 * have one.
501 */
Gideon Israel Dsouza3b321232014-04-07 15:37:26 -0700502void __weak vmalloc_sync_all(void)
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -0700503{
504}
505
Paul Mundt29c185e2010-12-24 12:08:30 +0900506/**
507 * alloc_vm_area - allocate a range of kernel address space
508 * @size: size of the area
509 *
510 * Returns: NULL on failure, vm_struct on success
511 *
512 * This function reserves a range of kernel address space, and
513 * allocates pagetables to map that range. No actual mappings
514 * are created. If the kernel address space is not shared
515 * between processes, it syncs the pagetable across all
516 * processes.
517 */
David Vrabelcd129092011-09-29 16:53:32 +0100518struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
Paul Mundt29c185e2010-12-24 12:08:30 +0900519{
520 BUG();
521 return NULL;
522}
523EXPORT_SYMBOL_GPL(alloc_vm_area);
524
525void free_vm_area(struct vm_struct *area)
526{
527 BUG();
528}
529EXPORT_SYMBOL_GPL(free_vm_area);
530
Paul Mundtb5073172007-07-21 04:37:25 -0700531int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
532 struct page *page)
533{
534 return -EINVAL;
535}
536EXPORT_SYMBOL(vm_insert_page);
537
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -0700538/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 * sys_brk() for the most part doesn't need the global kernel
540 * lock, except when an application is doing something nasty
541 * like trying to un-brk an area that has already been mapped
542 * to a regular file. in this case, the unmapping will need
543 * to invoke file system routines that need the global lock.
544 */
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100545SYSCALL_DEFINE1(brk, unsigned long, brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546{
547 struct mm_struct *mm = current->mm;
548
549 if (brk < mm->start_brk || brk > mm->context.end_brk)
550 return mm->brk;
551
552 if (mm->brk == brk)
553 return mm->brk;
554
555 /*
556 * Always allow shrinking brk
557 */
558 if (brk <= mm->brk) {
559 mm->brk = brk;
560 return brk;
561 }
562
563 /*
564 * Ok, looks good - let it rip.
565 */
Mike Frysingercfe79c02010-01-06 17:23:23 +0000566 flush_icache_range(mm->brk, brk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 return mm->brk = brk;
568}
569
David Howells8feae132009-01-08 12:04:47 +0000570/*
571 * initialise the VMA and region record slabs
572 */
573void __init mmap_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574{
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -0700575 int ret;
576
Tejun Heo908c7f12014-09-08 09:51:29 +0900577 ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -0700578 VM_BUG_ON(ret);
David Howells33e5d7692009-04-02 16:56:32 -0700579 vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC);
David Howells8feae132009-01-08 12:04:47 +0000580}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581
David Howells8feae132009-01-08 12:04:47 +0000582/*
583 * validate the region tree
584 * - the caller must hold the region lock
585 */
586#ifdef CONFIG_DEBUG_NOMMU_REGIONS
587static noinline void validate_nommu_regions(void)
588{
589 struct vm_region *region, *last;
590 struct rb_node *p, *lastp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591
David Howells8feae132009-01-08 12:04:47 +0000592 lastp = rb_first(&nommu_region_tree);
593 if (!lastp)
594 return;
595
596 last = rb_entry(lastp, struct vm_region, vm_rb);
David Howells33e5d7692009-04-02 16:56:32 -0700597 BUG_ON(unlikely(last->vm_end <= last->vm_start));
598 BUG_ON(unlikely(last->vm_top < last->vm_end));
David Howells8feae132009-01-08 12:04:47 +0000599
600 while ((p = rb_next(lastp))) {
601 region = rb_entry(p, struct vm_region, vm_rb);
602 last = rb_entry(lastp, struct vm_region, vm_rb);
603
David Howells33e5d7692009-04-02 16:56:32 -0700604 BUG_ON(unlikely(region->vm_end <= region->vm_start));
605 BUG_ON(unlikely(region->vm_top < region->vm_end));
606 BUG_ON(unlikely(region->vm_start < last->vm_top));
David Howells8feae132009-01-08 12:04:47 +0000607
608 lastp = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 }
610}
David Howells8feae132009-01-08 12:04:47 +0000611#else
David Howells33e5d7692009-04-02 16:56:32 -0700612static void validate_nommu_regions(void)
613{
614}
David Howells8feae132009-01-08 12:04:47 +0000615#endif
616
617/*
618 * add a region into the global tree
619 */
620static void add_nommu_region(struct vm_region *region)
621{
622 struct vm_region *pregion;
623 struct rb_node **p, *parent;
624
625 validate_nommu_regions();
626
David Howells8feae132009-01-08 12:04:47 +0000627 parent = NULL;
628 p = &nommu_region_tree.rb_node;
629 while (*p) {
630 parent = *p;
631 pregion = rb_entry(parent, struct vm_region, vm_rb);
632 if (region->vm_start < pregion->vm_start)
633 p = &(*p)->rb_left;
634 else if (region->vm_start > pregion->vm_start)
635 p = &(*p)->rb_right;
636 else if (pregion == region)
637 return;
638 else
639 BUG();
640 }
641
642 rb_link_node(&region->vm_rb, parent, p);
643 rb_insert_color(&region->vm_rb, &nommu_region_tree);
644
645 validate_nommu_regions();
646}
647
648/*
649 * delete a region from the global tree
650 */
651static void delete_nommu_region(struct vm_region *region)
652{
653 BUG_ON(!nommu_region_tree.rb_node);
654
655 validate_nommu_regions();
656 rb_erase(&region->vm_rb, &nommu_region_tree);
657 validate_nommu_regions();
658}
659
660/*
661 * free a contiguous series of pages
662 */
663static void free_page_series(unsigned long from, unsigned long to)
664{
665 for (; from < to; from += PAGE_SIZE) {
666 struct page *page = virt_to_page(from);
667
668 kdebug("- free %lx", from);
David Howells33e5d7692009-04-02 16:56:32 -0700669 atomic_long_dec(&mmap_pages_allocated);
David Howells8feae132009-01-08 12:04:47 +0000670 if (page_count(page) != 1)
David Howells33e5d7692009-04-02 16:56:32 -0700671 kdebug("free page %p: refcount not one: %d",
672 page, page_count(page));
David Howells8feae132009-01-08 12:04:47 +0000673 put_page(page);
674 }
675}
676
677/*
678 * release a reference to a region
David Howells33e5d7692009-04-02 16:56:32 -0700679 * - the caller must hold the region semaphore for writing, which this releases
Paul Mundtdd8632a2009-01-08 12:04:47 +0000680 * - the region may not have been added to the tree yet, in which case vm_top
David Howells8feae132009-01-08 12:04:47 +0000681 * will equal vm_start
682 */
683static void __put_nommu_region(struct vm_region *region)
684 __releases(nommu_region_sem)
685{
David Howells1e2ae592010-01-15 17:01:33 -0800686 kenter("%p{%d}", region, region->vm_usage);
David Howells8feae132009-01-08 12:04:47 +0000687
688 BUG_ON(!nommu_region_tree.rb_node);
689
David Howells1e2ae592010-01-15 17:01:33 -0800690 if (--region->vm_usage == 0) {
Paul Mundtdd8632a2009-01-08 12:04:47 +0000691 if (region->vm_top > region->vm_start)
David Howells8feae132009-01-08 12:04:47 +0000692 delete_nommu_region(region);
693 up_write(&nommu_region_sem);
694
695 if (region->vm_file)
696 fput(region->vm_file);
697
698 /* IO memory and memory shared directly out of the pagecache
699 * from ramfs/tmpfs mustn't be released here */
700 if (region->vm_flags & VM_MAPPED_COPY) {
701 kdebug("free series");
Paul Mundtdd8632a2009-01-08 12:04:47 +0000702 free_page_series(region->vm_start, region->vm_top);
David Howells8feae132009-01-08 12:04:47 +0000703 }
704 kmem_cache_free(vm_region_jar, region);
705 } else {
706 up_write(&nommu_region_sem);
707 }
708}
709
710/*
711 * release a reference to a region
712 */
713static void put_nommu_region(struct vm_region *region)
714{
715 down_write(&nommu_region_sem);
716 __put_nommu_region(region);
717}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718
David Howells30340972006-09-27 01:50:20 -0700719/*
Bernd Schmidteb8cdec2009-09-21 17:03:57 -0700720 * update protection on a vma
721 */
722static void protect_vma(struct vm_area_struct *vma, unsigned long flags)
723{
724#ifdef CONFIG_MPU
725 struct mm_struct *mm = vma->vm_mm;
726 long start = vma->vm_start & PAGE_MASK;
727 while (start < vma->vm_end) {
728 protect_page(mm, start, flags);
729 start += PAGE_SIZE;
730 }
731 update_protections(mm);
732#endif
733}
734
735/*
David Howells30340972006-09-27 01:50:20 -0700736 * add a VMA into a process's mm_struct in the appropriate place in the list
David Howells8feae132009-01-08 12:04:47 +0000737 * and tree and add to the address space's page tree also if not an anonymous
738 * page
David Howells30340972006-09-27 01:50:20 -0700739 * - should be called with mm->mmap_sem held writelocked
740 */
David Howells8feae132009-01-08 12:04:47 +0000741static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
David Howells30340972006-09-27 01:50:20 -0700742{
Namhyung Kim6038def2011-05-24 17:11:22 -0700743 struct vm_area_struct *pvma, *prev;
David Howells8feae132009-01-08 12:04:47 +0000744 struct address_space *mapping;
Namhyung Kim6038def2011-05-24 17:11:22 -0700745 struct rb_node **p, *parent, *rb_prev;
David Howells30340972006-09-27 01:50:20 -0700746
David Howells8feae132009-01-08 12:04:47 +0000747 kenter(",%p", vma);
748
749 BUG_ON(!vma->vm_region);
750
751 mm->map_count++;
752 vma->vm_mm = mm;
753
Bernd Schmidteb8cdec2009-09-21 17:03:57 -0700754 protect_vma(vma, vma->vm_flags);
755
David Howells8feae132009-01-08 12:04:47 +0000756 /* add the VMA to the mapping */
757 if (vma->vm_file) {
758 mapping = vma->vm_file->f_mapping;
759
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800760 i_mmap_lock_write(mapping);
David Howells8feae132009-01-08 12:04:47 +0000761 flush_dcache_mmap_lock(mapping);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700762 vma_interval_tree_insert(vma, &mapping->i_mmap);
David Howells8feae132009-01-08 12:04:47 +0000763 flush_dcache_mmap_unlock(mapping);
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800764 i_mmap_unlock_write(mapping);
David Howells8feae132009-01-08 12:04:47 +0000765 }
766
767 /* add the VMA to the tree */
Namhyung Kim6038def2011-05-24 17:11:22 -0700768 parent = rb_prev = NULL;
David Howells8feae132009-01-08 12:04:47 +0000769 p = &mm->mm_rb.rb_node;
770 while (*p) {
771 parent = *p;
772 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
773
774 /* sort by: start addr, end addr, VMA struct addr in that order
775 * (the latter is necessary as we may get identical VMAs) */
776 if (vma->vm_start < pvma->vm_start)
777 p = &(*p)->rb_left;
Namhyung Kim6038def2011-05-24 17:11:22 -0700778 else if (vma->vm_start > pvma->vm_start) {
779 rb_prev = parent;
David Howells8feae132009-01-08 12:04:47 +0000780 p = &(*p)->rb_right;
Namhyung Kim6038def2011-05-24 17:11:22 -0700781 } else if (vma->vm_end < pvma->vm_end)
David Howells8feae132009-01-08 12:04:47 +0000782 p = &(*p)->rb_left;
Namhyung Kim6038def2011-05-24 17:11:22 -0700783 else if (vma->vm_end > pvma->vm_end) {
784 rb_prev = parent;
David Howells8feae132009-01-08 12:04:47 +0000785 p = &(*p)->rb_right;
Namhyung Kim6038def2011-05-24 17:11:22 -0700786 } else if (vma < pvma)
David Howells8feae132009-01-08 12:04:47 +0000787 p = &(*p)->rb_left;
Namhyung Kim6038def2011-05-24 17:11:22 -0700788 else if (vma > pvma) {
789 rb_prev = parent;
David Howells8feae132009-01-08 12:04:47 +0000790 p = &(*p)->rb_right;
Namhyung Kim6038def2011-05-24 17:11:22 -0700791 } else
David Howells8feae132009-01-08 12:04:47 +0000792 BUG();
793 }
794
795 rb_link_node(&vma->vm_rb, parent, p);
796 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
797
798 /* add VMA to the VMA list also */
Namhyung Kim6038def2011-05-24 17:11:22 -0700799 prev = NULL;
800 if (rb_prev)
801 prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
David Howells30340972006-09-27 01:50:20 -0700802
Namhyung Kim6038def2011-05-24 17:11:22 -0700803 __vma_link_list(mm, vma, prev, parent);
David Howells8feae132009-01-08 12:04:47 +0000804}
805
806/*
807 * delete a VMA from its owning mm_struct and address space
808 */
809static void delete_vma_from_mm(struct vm_area_struct *vma)
810{
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700811 int i;
David Howells8feae132009-01-08 12:04:47 +0000812 struct address_space *mapping;
813 struct mm_struct *mm = vma->vm_mm;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700814 struct task_struct *curr = current;
David Howells8feae132009-01-08 12:04:47 +0000815
816 kenter("%p", vma);
817
Bernd Schmidteb8cdec2009-09-21 17:03:57 -0700818 protect_vma(vma, 0);
819
David Howells8feae132009-01-08 12:04:47 +0000820 mm->map_count--;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700821 for (i = 0; i < VMACACHE_SIZE; i++) {
822 /* if the vma is cached, invalidate the entire cache */
823 if (curr->vmacache[i] == vma) {
Steven Miaoe020d5b2014-06-23 13:22:02 -0700824 vmacache_invalidate(mm);
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700825 break;
826 }
827 }
David Howells8feae132009-01-08 12:04:47 +0000828
829 /* remove the VMA from the mapping */
830 if (vma->vm_file) {
831 mapping = vma->vm_file->f_mapping;
832
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800833 i_mmap_lock_write(mapping);
David Howells8feae132009-01-08 12:04:47 +0000834 flush_dcache_mmap_lock(mapping);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700835 vma_interval_tree_remove(vma, &mapping->i_mmap);
David Howells8feae132009-01-08 12:04:47 +0000836 flush_dcache_mmap_unlock(mapping);
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800837 i_mmap_unlock_write(mapping);
David Howells8feae132009-01-08 12:04:47 +0000838 }
839
840 /* remove from the MM's tree and list */
841 rb_erase(&vma->vm_rb, &mm->mm_rb);
Namhyung Kimb951bf22011-05-24 17:11:23 -0700842
843 if (vma->vm_prev)
844 vma->vm_prev->vm_next = vma->vm_next;
845 else
846 mm->mmap = vma->vm_next;
847
848 if (vma->vm_next)
849 vma->vm_next->vm_prev = vma->vm_prev;
David Howells8feae132009-01-08 12:04:47 +0000850}
851
852/*
853 * destroy a VMA record
854 */
855static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
856{
857 kenter("%p", vma);
858 if (vma->vm_ops && vma->vm_ops->close)
859 vma->vm_ops->close(vma);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -0700860 if (vma->vm_file)
David Howells8feae132009-01-08 12:04:47 +0000861 fput(vma->vm_file);
David Howells8feae132009-01-08 12:04:47 +0000862 put_nommu_region(vma->vm_region);
863 kmem_cache_free(vm_area_cachep, vma);
David Howells30340972006-09-27 01:50:20 -0700864}
865
866/*
867 * look up the first VMA in which addr resides, NULL if none
868 * - should be called with mm->mmap_sem at least held readlocked
869 */
870struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
871{
David Howells8feae132009-01-08 12:04:47 +0000872 struct vm_area_struct *vma;
David Howells30340972006-09-27 01:50:20 -0700873
David Howells8feae132009-01-08 12:04:47 +0000874 /* check the cache first */
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700875 vma = vmacache_find(mm, addr);
876 if (likely(vma))
David Howells8feae132009-01-08 12:04:47 +0000877 return vma;
878
Namhyung Kime922c4c2011-05-24 17:11:24 -0700879 /* trawl the list (there may be multiple mappings in which addr
David Howells8feae132009-01-08 12:04:47 +0000880 * resides) */
Namhyung Kime922c4c2011-05-24 17:11:24 -0700881 for (vma = mm->mmap; vma; vma = vma->vm_next) {
David Howells8feae132009-01-08 12:04:47 +0000882 if (vma->vm_start > addr)
883 return NULL;
884 if (vma->vm_end > addr) {
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700885 vmacache_update(addr, vma);
David Howells8feae132009-01-08 12:04:47 +0000886 return vma;
887 }
David Howells30340972006-09-27 01:50:20 -0700888 }
889
David Howells30340972006-09-27 01:50:20 -0700890 return NULL;
891}
892EXPORT_SYMBOL(find_vma);
893
894/*
David Howells930e6522006-09-27 01:50:22 -0700895 * find a VMA
896 * - we don't extend stack VMAs under NOMMU conditions
897 */
898struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
899{
David Howells7561e8c2010-03-25 16:48:38 +0000900 return find_vma(mm, addr);
David Howells930e6522006-09-27 01:50:22 -0700901}
902
David Howells8feae132009-01-08 12:04:47 +0000903/*
904 * expand a stack to a given address
905 * - not supported under NOMMU conditions
906 */
Greg Ungerer57c8f632007-07-15 23:38:28 -0700907int expand_stack(struct vm_area_struct *vma, unsigned long address)
908{
909 return -ENOMEM;
910}
911
David Howells930e6522006-09-27 01:50:22 -0700912/*
David Howells6fa5f802006-09-27 01:50:21 -0700913 * look up the first VMA exactly that exactly matches addr
914 * - should be called with mm->mmap_sem at least held readlocked
915 */
David Howells8feae132009-01-08 12:04:47 +0000916static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
917 unsigned long addr,
918 unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919{
920 struct vm_area_struct *vma;
David Howells8feae132009-01-08 12:04:47 +0000921 unsigned long end = addr + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922
David Howells8feae132009-01-08 12:04:47 +0000923 /* check the cache first */
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700924 vma = vmacache_find_exact(mm, addr, end);
925 if (vma)
David Howells8feae132009-01-08 12:04:47 +0000926 return vma;
927
Namhyung Kime922c4c2011-05-24 17:11:24 -0700928 /* trawl the list (there may be multiple mappings in which addr
David Howells8feae132009-01-08 12:04:47 +0000929 * resides) */
Namhyung Kime922c4c2011-05-24 17:11:24 -0700930 for (vma = mm->mmap; vma; vma = vma->vm_next) {
David Howells8feae132009-01-08 12:04:47 +0000931 if (vma->vm_start < addr)
932 continue;
933 if (vma->vm_start > addr)
934 return NULL;
935 if (vma->vm_end == end) {
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700936 vmacache_update(addr, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 return vma;
David Howells8feae132009-01-08 12:04:47 +0000938 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 }
940
941 return NULL;
942}
943
David Howells30340972006-09-27 01:50:20 -0700944/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 * determine whether a mapping should be permitted and, if so, what sort of
946 * mapping we're capable of supporting
947 */
948static int validate_mmap_request(struct file *file,
949 unsigned long addr,
950 unsigned long len,
951 unsigned long prot,
952 unsigned long flags,
953 unsigned long pgoff,
954 unsigned long *_capabilities)
955{
David Howells8feae132009-01-08 12:04:47 +0000956 unsigned long capabilities, rlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 int ret;
958
959 /* do the simple checks first */
David Howells06aab5a2009-09-24 12:33:48 +0100960 if (flags & MAP_FIXED) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 printk(KERN_DEBUG
962 "%d: Can't do fixed-address/overlay mmap of RAM\n",
963 current->pid);
964 return -EINVAL;
965 }
966
967 if ((flags & MAP_TYPE) != MAP_PRIVATE &&
968 (flags & MAP_TYPE) != MAP_SHARED)
969 return -EINVAL;
970
Mike Frysingerf81cff0d2006-12-06 12:02:59 +1000971 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 return -EINVAL;
973
Mike Frysingerf81cff0d2006-12-06 12:02:59 +1000974 /* Careful about overflows.. */
David Howells8feae132009-01-08 12:04:47 +0000975 rlen = PAGE_ALIGN(len);
976 if (!rlen || rlen > TASK_SIZE)
Mike Frysingerf81cff0d2006-12-06 12:02:59 +1000977 return -ENOMEM;
978
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 /* offset overflow? */
David Howells8feae132009-01-08 12:04:47 +0000980 if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
Mike Frysingerf81cff0d2006-12-06 12:02:59 +1000981 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982
983 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 /* files must support mmap */
Al Viro72c2d532013-09-22 16:27:52 -0400985 if (!file->f_op->mmap)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 return -ENODEV;
987
988 /* work out if what we've got could possibly be shared
989 * - we support chardevs that provide their own "memory"
990 * - we support files/blockdevs that are memory backed
991 */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100992 if (file->f_op->mmap_capabilities) {
993 capabilities = file->f_op->mmap_capabilities(file);
994 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 /* no explicit capabilities set, so assume some
996 * defaults */
Al Viro496ad9a2013-01-23 17:07:38 -0500997 switch (file_inode(file)->i_mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998 case S_IFREG:
999 case S_IFBLK:
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001000 capabilities = NOMMU_MAP_COPY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 break;
1002
1003 case S_IFCHR:
1004 capabilities =
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001005 NOMMU_MAP_DIRECT |
1006 NOMMU_MAP_READ |
1007 NOMMU_MAP_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 break;
1009
1010 default:
1011 return -EINVAL;
1012 }
1013 }
1014
1015 /* eliminate any capabilities that we can't support on this
1016 * device */
1017 if (!file->f_op->get_unmapped_area)
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001018 capabilities &= ~NOMMU_MAP_DIRECT;
Al Viro6e242a12015-03-31 12:35:13 -04001019 if (!(file->f_mode & FMODE_CAN_READ))
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001020 capabilities &= ~NOMMU_MAP_COPY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021
Graff Yang28d7a6a2009-08-18 14:11:17 -07001022 /* The file shall have been opened with read permission. */
1023 if (!(file->f_mode & FMODE_READ))
1024 return -EACCES;
1025
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 if (flags & MAP_SHARED) {
1027 /* do checks for writing, appending and locking */
1028 if ((prot & PROT_WRITE) &&
1029 !(file->f_mode & FMODE_WRITE))
1030 return -EACCES;
1031
Al Viro496ad9a2013-01-23 17:07:38 -05001032 if (IS_APPEND(file_inode(file)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 (file->f_mode & FMODE_WRITE))
1034 return -EACCES;
1035
Jeff Laytond7a06982014-03-10 09:54:15 -04001036 if (locks_verify_locked(file))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 return -EAGAIN;
1038
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001039 if (!(capabilities & NOMMU_MAP_DIRECT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 return -ENODEV;
1041
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 /* we mustn't privatise shared mappings */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001043 capabilities &= ~NOMMU_MAP_COPY;
Choi Gi-yongac714902014-04-07 15:37:36 -07001044 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 /* we're going to read the file into private memory we
1046 * allocate */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001047 if (!(capabilities & NOMMU_MAP_COPY))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 return -ENODEV;
1049
1050 /* we don't permit a private writable mapping to be
1051 * shared with the backing device */
1052 if (prot & PROT_WRITE)
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001053 capabilities &= ~NOMMU_MAP_DIRECT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 }
1055
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001056 if (capabilities & NOMMU_MAP_DIRECT) {
1057 if (((prot & PROT_READ) && !(capabilities & NOMMU_MAP_READ)) ||
1058 ((prot & PROT_WRITE) && !(capabilities & NOMMU_MAP_WRITE)) ||
1059 ((prot & PROT_EXEC) && !(capabilities & NOMMU_MAP_EXEC))
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001060 ) {
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001061 capabilities &= ~NOMMU_MAP_DIRECT;
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001062 if (flags & MAP_SHARED) {
1063 printk(KERN_WARNING
1064 "MAP_SHARED not completely supported on !MMU\n");
1065 return -EINVAL;
1066 }
1067 }
1068 }
1069
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 /* handle executable mappings and implied executable
1071 * mappings */
Josef Sipeke9536ae2006-12-08 02:37:21 -08001072 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 if (prot & PROT_EXEC)
1074 return -EPERM;
Choi Gi-yongac714902014-04-07 15:37:36 -07001075 } else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 /* handle implication of PROT_EXEC by PROT_READ */
1077 if (current->personality & READ_IMPLIES_EXEC) {
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001078 if (capabilities & NOMMU_MAP_EXEC)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 prot |= PROT_EXEC;
1080 }
Choi Gi-yongac714902014-04-07 15:37:36 -07001081 } else if ((prot & PROT_READ) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 (prot & PROT_EXEC) &&
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001083 !(capabilities & NOMMU_MAP_EXEC)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 ) {
1085 /* backing file is not executable, try to copy */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001086 capabilities &= ~NOMMU_MAP_DIRECT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 }
Choi Gi-yongac714902014-04-07 15:37:36 -07001088 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089 /* anonymous mappings are always memory backed and can be
1090 * privately mapped
1091 */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001092 capabilities = NOMMU_MAP_COPY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093
1094 /* handle PROT_EXEC implication by PROT_READ */
1095 if ((prot & PROT_READ) &&
1096 (current->personality & READ_IMPLIES_EXEC))
1097 prot |= PROT_EXEC;
1098 }
1099
1100 /* allow the security API to have its say */
Al Viroe5467852012-05-30 13:30:51 -04001101 ret = security_mmap_addr(addr);
1102 if (ret < 0)
1103 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104
1105 /* looks okay */
1106 *_capabilities = capabilities;
1107 return 0;
1108}
1109
1110/*
1111 * we've determined that we can make the mapping, now translate what we
1112 * now know into VMA flags
1113 */
1114static unsigned long determine_vm_flags(struct file *file,
1115 unsigned long prot,
1116 unsigned long flags,
1117 unsigned long capabilities)
1118{
1119 unsigned long vm_flags;
1120
1121 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 /* vm_flags |= mm->def_flags; */
1123
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001124 if (!(capabilities & NOMMU_MAP_DIRECT)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 /* attempt to share read-only copies of mapped file chunks */
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001126 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 if (file && !(prot & PROT_WRITE))
1128 vm_flags |= VM_MAYSHARE;
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001129 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 /* overlay a shareable mapping on the backing device or inode
1131 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1132 * romfs/cramfs */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001133 vm_flags |= VM_MAYSHARE | (capabilities & NOMMU_VMFLAGS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134 if (flags & MAP_SHARED)
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001135 vm_flags |= VM_SHARED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 }
1137
1138 /* refuse to let anyone share private mappings with this process if
1139 * it's being traced - otherwise breakpoints set in it may interfere
1140 * with another untraced process
1141 */
Tejun Heoa288eec2011-06-17 16:50:37 +02001142 if ((flags & MAP_PRIVATE) && current->ptrace)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 vm_flags &= ~VM_MAYSHARE;
1144
1145 return vm_flags;
1146}
1147
1148/*
David Howells8feae132009-01-08 12:04:47 +00001149 * set up a shared mapping on a file (the driver or filesystem provides and
1150 * pins the storage)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151 */
David Howells8feae132009-01-08 12:04:47 +00001152static int do_mmap_shared_file(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153{
1154 int ret;
1155
1156 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001157 if (ret == 0) {
1158 vma->vm_region->vm_top = vma->vm_region->vm_end;
David Howells645d83c2009-09-24 15:13:10 +01001159 return 0;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001160 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 if (ret != -ENOSYS)
1162 return ret;
1163
David Howells3fa30462010-03-23 13:35:21 -07001164 /* getting -ENOSYS indicates that direct mmap isn't possible (as
1165 * opposed to tried but failed) so we can only give a suitable error as
1166 * it's not possible to make a private copy if MAP_SHARED was given */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167 return -ENODEV;
1168}
1169
1170/*
1171 * set up a private mapping or an anonymous shared mapping
1172 */
David Howells8feae132009-01-08 12:04:47 +00001173static int do_mmap_private(struct vm_area_struct *vma,
1174 struct vm_region *region,
David Howells645d83c2009-09-24 15:13:10 +01001175 unsigned long len,
1176 unsigned long capabilities)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177{
Joonsoo Kimdbc83582014-12-12 16:55:55 -08001178 unsigned long total, point;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 void *base;
David Howells8feae132009-01-08 12:04:47 +00001180 int ret, order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181
1182 /* invoke the file's mapping function so that it can keep track of
1183 * shared mappings on devices or memory
1184 * - VM_MAYSHARE will be set if it may attempt to share
1185 */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001186 if (capabilities & NOMMU_MAP_DIRECT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001188 if (ret == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189 /* shouldn't return success if we're not sharing */
Paul Mundtdd8632a2009-01-08 12:04:47 +00001190 BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1191 vma->vm_region->vm_top = vma->vm_region->vm_end;
David Howells645d83c2009-09-24 15:13:10 +01001192 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193 }
Paul Mundtdd8632a2009-01-08 12:04:47 +00001194 if (ret != -ENOSYS)
1195 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196
1197 /* getting an ENOSYS error indicates that direct mmap isn't
1198 * possible (as opposed to tried but failed) so we'll try to
1199 * make a private copy of the data and map that instead */
1200 }
1201
David Howells8feae132009-01-08 12:04:47 +00001202
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 /* allocate some memory to hold the mapping
1204 * - note that this may not return a page-aligned address if the object
1205 * we're allocating is smaller than a page
1206 */
Bob Liuf67d9b12011-05-24 17:12:56 -07001207 order = get_order(len);
David Howells8feae132009-01-08 12:04:47 +00001208 kdebug("alloc order %d for %lx", order, len);
1209
David Howells8feae132009-01-08 12:04:47 +00001210 total = 1 << order;
Bob Liuf67d9b12011-05-24 17:12:56 -07001211 point = len >> PAGE_SHIFT;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001212
Joonsoo Kimdbc83582014-12-12 16:55:55 -08001213 /* we don't want to allocate a power-of-2 sized page set */
Paul Mundtdd8632a2009-01-08 12:04:47 +00001214 if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) {
Joonsoo Kimdbc83582014-12-12 16:55:55 -08001215 total = point;
1216 kdebug("try to alloc exact %lu pages", total);
David Howells8feae132009-01-08 12:04:47 +00001217 }
1218
Joonsoo Kimda616532015-02-27 15:51:43 -08001219 base = alloc_pages_exact(total << PAGE_SHIFT, GFP_KERNEL);
Joonsoo Kimdbc83582014-12-12 16:55:55 -08001220 if (!base)
1221 goto enomem;
David Howells8feae132009-01-08 12:04:47 +00001222
Joonsoo Kimdbc83582014-12-12 16:55:55 -08001223 atomic_long_add(total, &mmap_pages_allocated);
1224
David Howells8feae132009-01-08 12:04:47 +00001225 region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1226 region->vm_start = (unsigned long) base;
Bob Liuf67d9b12011-05-24 17:12:56 -07001227 region->vm_end = region->vm_start + len;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001228 region->vm_top = region->vm_start + (total << PAGE_SHIFT);
David Howells8feae132009-01-08 12:04:47 +00001229
1230 vma->vm_start = region->vm_start;
1231 vma->vm_end = region->vm_start + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232
1233 if (vma->vm_file) {
1234 /* read the contents of a file into the copy */
1235 mm_segment_t old_fs;
1236 loff_t fpos;
1237
1238 fpos = vma->vm_pgoff;
1239 fpos <<= PAGE_SHIFT;
1240
1241 old_fs = get_fs();
1242 set_fs(KERNEL_DS);
Al Viro6e242a12015-03-31 12:35:13 -04001243 ret = __vfs_read(vma->vm_file, base, len, &fpos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 set_fs(old_fs);
1245
1246 if (ret < 0)
1247 goto error_free;
1248
1249 /* clear the last little bit */
Bob Liuf67d9b12011-05-24 17:12:56 -07001250 if (ret < len)
1251 memset(base + ret, 0, len - ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 }
1254
1255 return 0;
1256
1257error_free:
Namhyung Kim7223bb42011-05-24 17:11:26 -07001258 free_page_series(region->vm_start, region->vm_top);
David Howells8feae132009-01-08 12:04:47 +00001259 region->vm_start = vma->vm_start = 0;
1260 region->vm_end = vma->vm_end = 0;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001261 region->vm_top = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 return ret;
1263
1264enomem:
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -07001265 pr_err("Allocation of length %lu from process %d (%s) failed\n",
Greg Ungerer05ae6fa2009-01-13 17:30:22 +10001266 len, current->pid, current->comm);
David Rientjes7bf02ea2011-05-24 17:11:16 -07001267 show_free_areas(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 return -ENOMEM;
1269}
1270
1271/*
1272 * handle mapping creation for uClinux
1273 */
Al Viroe3fc6292012-05-30 20:08:42 -04001274unsigned long do_mmap_pgoff(struct file *file,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275 unsigned long addr,
1276 unsigned long len,
1277 unsigned long prot,
1278 unsigned long flags,
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001279 unsigned long pgoff,
Michel Lespinasse41badc12013-02-22 16:32:47 -08001280 unsigned long *populate)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281{
David Howells8feae132009-01-08 12:04:47 +00001282 struct vm_area_struct *vma;
1283 struct vm_region *region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 struct rb_node *rb;
David Howells8feae132009-01-08 12:04:47 +00001285 unsigned long capabilities, vm_flags, result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 int ret;
1287
David Howells8feae132009-01-08 12:04:47 +00001288 kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff);
1289
Michel Lespinasse41badc12013-02-22 16:32:47 -08001290 *populate = 0;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001291
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292 /* decide whether we should attempt the mapping, and if so what sort of
1293 * mapping */
1294 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1295 &capabilities);
David Howells8feae132009-01-08 12:04:47 +00001296 if (ret < 0) {
1297 kleave(" = %d [val]", ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 return ret;
David Howells8feae132009-01-08 12:04:47 +00001299 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300
David Howells06aab5a2009-09-24 12:33:48 +01001301 /* we ignore the address hint */
1302 addr = 0;
Bob Liuf67d9b12011-05-24 17:12:56 -07001303 len = PAGE_ALIGN(len);
David Howells06aab5a2009-09-24 12:33:48 +01001304
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 /* we've determined that we can make the mapping, now translate what we
1306 * now know into VMA flags */
1307 vm_flags = determine_vm_flags(file, prot, flags, capabilities);
1308
David Howells8feae132009-01-08 12:04:47 +00001309 /* we're going to need to record the mapping */
1310 region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1311 if (!region)
1312 goto error_getting_region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313
David Howells8feae132009-01-08 12:04:47 +00001314 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1315 if (!vma)
1316 goto error_getting_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317
David Howells1e2ae592010-01-15 17:01:33 -08001318 region->vm_usage = 1;
David Howells8feae132009-01-08 12:04:47 +00001319 region->vm_flags = vm_flags;
1320 region->vm_pgoff = pgoff;
1321
Rik van Riel5beb4932010-03-05 13:42:07 -08001322 INIT_LIST_HEAD(&vma->anon_vma_chain);
David Howells8feae132009-01-08 12:04:47 +00001323 vma->vm_flags = vm_flags;
1324 vma->vm_pgoff = pgoff;
1325
1326 if (file) {
Al Virocb0942b2012-08-27 14:48:26 -04001327 region->vm_file = get_file(file);
1328 vma->vm_file = get_file(file);
David Howells8feae132009-01-08 12:04:47 +00001329 }
1330
1331 down_write(&nommu_region_sem);
1332
1333 /* if we want to share, we need to check for regions created by other
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334 * mmap() calls that overlap with our proposed mapping
David Howells8feae132009-01-08 12:04:47 +00001335 * - we can only share with a superset match on most regular files
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 * - shared mappings on character devices and memory backed files are
1337 * permitted to overlap inexactly as far as we are concerned for in
1338 * these cases, sharing is handled in the driver or filesystem rather
1339 * than here
1340 */
1341 if (vm_flags & VM_MAYSHARE) {
David Howells8feae132009-01-08 12:04:47 +00001342 struct vm_region *pregion;
1343 unsigned long pglen, rpglen, pgend, rpgend, start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344
David Howells8feae132009-01-08 12:04:47 +00001345 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1346 pgend = pgoff + pglen;
David Howells165b2392007-03-22 00:11:24 -08001347
David Howells8feae132009-01-08 12:04:47 +00001348 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1349 pregion = rb_entry(rb, struct vm_region, vm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350
David Howells8feae132009-01-08 12:04:47 +00001351 if (!(pregion->vm_flags & VM_MAYSHARE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 continue;
1353
1354 /* search for overlapping mappings on the same file */
Al Viro496ad9a2013-01-23 17:07:38 -05001355 if (file_inode(pregion->vm_file) !=
1356 file_inode(file))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 continue;
1358
David Howells8feae132009-01-08 12:04:47 +00001359 if (pregion->vm_pgoff >= pgend)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360 continue;
1361
David Howells8feae132009-01-08 12:04:47 +00001362 rpglen = pregion->vm_end - pregion->vm_start;
1363 rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1364 rpgend = pregion->vm_pgoff + rpglen;
1365 if (pgoff >= rpgend)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 continue;
1367
David Howells8feae132009-01-08 12:04:47 +00001368 /* handle inexactly overlapping matches between
1369 * mappings */
1370 if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1371 !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1372 /* new mapping is not a subset of the region */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001373 if (!(capabilities & NOMMU_MAP_DIRECT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 goto sharing_violation;
1375 continue;
1376 }
1377
David Howells8feae132009-01-08 12:04:47 +00001378 /* we've found a region we can share */
David Howells1e2ae592010-01-15 17:01:33 -08001379 pregion->vm_usage++;
David Howells8feae132009-01-08 12:04:47 +00001380 vma->vm_region = pregion;
1381 start = pregion->vm_start;
1382 start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1383 vma->vm_start = start;
1384 vma->vm_end = start + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385
David Howells8feae132009-01-08 12:04:47 +00001386 if (pregion->vm_flags & VM_MAPPED_COPY) {
1387 kdebug("share copy");
1388 vma->vm_flags |= VM_MAPPED_COPY;
1389 } else {
1390 kdebug("share mmap");
1391 ret = do_mmap_shared_file(vma);
1392 if (ret < 0) {
1393 vma->vm_region = NULL;
1394 vma->vm_start = 0;
1395 vma->vm_end = 0;
David Howells1e2ae592010-01-15 17:01:33 -08001396 pregion->vm_usage--;
David Howells8feae132009-01-08 12:04:47 +00001397 pregion = NULL;
1398 goto error_just_free;
1399 }
1400 }
1401 fput(region->vm_file);
1402 kmem_cache_free(vm_region_jar, region);
1403 region = pregion;
1404 result = start;
1405 goto share;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 }
1407
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 /* obtain the address at which to make a shared mapping
1409 * - this is the hook for quasi-memory character devices to
1410 * tell us the location of a shared mapping
1411 */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001412 if (capabilities & NOMMU_MAP_DIRECT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 addr = file->f_op->get_unmapped_area(file, addr, len,
1414 pgoff, flags);
Namhyung Kimbb005a52011-05-24 17:11:27 -07001415 if (IS_ERR_VALUE(addr)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416 ret = addr;
Namhyung Kimbb005a52011-05-24 17:11:27 -07001417 if (ret != -ENOSYS)
David Howells8feae132009-01-08 12:04:47 +00001418 goto error_just_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419
1420 /* the driver refused to tell us where to site
1421 * the mapping so we'll have to attempt to copy
1422 * it */
Namhyung Kimbb005a52011-05-24 17:11:27 -07001423 ret = -ENODEV;
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001424 if (!(capabilities & NOMMU_MAP_COPY))
David Howells8feae132009-01-08 12:04:47 +00001425 goto error_just_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001427 capabilities &= ~NOMMU_MAP_DIRECT;
David Howells8feae132009-01-08 12:04:47 +00001428 } else {
1429 vma->vm_start = region->vm_start = addr;
1430 vma->vm_end = region->vm_end = addr + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431 }
1432 }
1433 }
1434
David Howells8feae132009-01-08 12:04:47 +00001435 vma->vm_region = region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436
David Howells645d83c2009-09-24 15:13:10 +01001437 /* set up the mapping
Christoph Hellwigb4caecd2015-01-14 10:42:32 +01001438 * - the region is filled in if NOMMU_MAP_DIRECT is still set
David Howells645d83c2009-09-24 15:13:10 +01001439 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 if (file && vma->vm_flags & VM_SHARED)
David Howells8feae132009-01-08 12:04:47 +00001441 ret = do_mmap_shared_file(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442 else
David Howells645d83c2009-09-24 15:13:10 +01001443 ret = do_mmap_private(vma, region, len, capabilities);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444 if (ret < 0)
David Howells645d83c2009-09-24 15:13:10 +01001445 goto error_just_free;
1446 add_nommu_region(region);
David Howells8feae132009-01-08 12:04:47 +00001447
Jie Zhangea637632009-12-14 18:00:02 -08001448 /* clear anonymous mappings that don't ask for uninitialized data */
1449 if (!vma->vm_file && !(flags & MAP_UNINITIALIZED))
1450 memset((void *)region->vm_start, 0,
1451 region->vm_end - region->vm_start);
1452
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 /* okay... we have a mapping; now we have to register it */
David Howells8feae132009-01-08 12:04:47 +00001454 result = vma->vm_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456 current->mm->total_vm += len >> PAGE_SHIFT;
1457
David Howells8feae132009-01-08 12:04:47 +00001458share:
1459 add_vma_to_mm(current->mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460
Mike Frysingercfe79c02010-01-06 17:23:23 +00001461 /* we flush the region from the icache only when the first executable
1462 * mapping of it is made */
1463 if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1464 flush_icache_range(region->vm_start, region->vm_end);
1465 region->vm_icache_flushed = true;
1466 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467
Mike Frysingercfe79c02010-01-06 17:23:23 +00001468 up_write(&nommu_region_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469
David Howells8feae132009-01-08 12:04:47 +00001470 kleave(" = %lx", result);
1471 return result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472
David Howells8feae132009-01-08 12:04:47 +00001473error_just_free:
1474 up_write(&nommu_region_sem);
1475error:
David Howells89a86402009-10-30 13:13:26 +00001476 if (region->vm_file)
1477 fput(region->vm_file);
David Howells8feae132009-01-08 12:04:47 +00001478 kmem_cache_free(vm_region_jar, region);
David Howells89a86402009-10-30 13:13:26 +00001479 if (vma->vm_file)
1480 fput(vma->vm_file);
David Howells8feae132009-01-08 12:04:47 +00001481 kmem_cache_free(vm_area_cachep, vma);
1482 kleave(" = %d", ret);
1483 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484
David Howells8feae132009-01-08 12:04:47 +00001485sharing_violation:
1486 up_write(&nommu_region_sem);
1487 printk(KERN_WARNING "Attempt to share mismatched mappings\n");
1488 ret = -EINVAL;
1489 goto error;
1490
1491error_getting_vma:
1492 kmem_cache_free(vm_region_jar, region);
1493 printk(KERN_WARNING "Allocation of vma for %lu byte allocation"
1494 " from process %d failed\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495 len, current->pid);
David Rientjes7bf02ea2011-05-24 17:11:16 -07001496 show_free_areas(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 return -ENOMEM;
1498
David Howells8feae132009-01-08 12:04:47 +00001499error_getting_region:
1500 printk(KERN_WARNING "Allocation of vm region for %lu byte allocation"
1501 " from process %d failed\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 len, current->pid);
David Rientjes7bf02ea2011-05-24 17:11:16 -07001503 show_free_areas(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 return -ENOMEM;
1505}
Linus Torvalds6be5ceb2012-04-20 17:13:58 -07001506
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001507SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1508 unsigned long, prot, unsigned long, flags,
1509 unsigned long, fd, unsigned long, pgoff)
1510{
1511 struct file *file = NULL;
1512 unsigned long retval = -EBADF;
1513
Al Viro120a7952010-10-30 02:54:44 -04001514 audit_mmap_fd(fd, flags);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001515 if (!(flags & MAP_ANONYMOUS)) {
1516 file = fget(fd);
1517 if (!file)
1518 goto out;
1519 }
1520
1521 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1522
Greg Ungererad1ed292012-06-04 14:29:59 +10001523 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001524
1525 if (file)
1526 fput(file);
1527out:
1528 return retval;
1529}
1530
Christoph Hellwiga4679372010-03-10 15:21:15 -08001531#ifdef __ARCH_WANT_SYS_OLD_MMAP
1532struct mmap_arg_struct {
1533 unsigned long addr;
1534 unsigned long len;
1535 unsigned long prot;
1536 unsigned long flags;
1537 unsigned long fd;
1538 unsigned long offset;
1539};
1540
1541SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1542{
1543 struct mmap_arg_struct a;
1544
1545 if (copy_from_user(&a, arg, sizeof(a)))
1546 return -EFAULT;
1547 if (a.offset & ~PAGE_MASK)
1548 return -EINVAL;
1549
1550 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1551 a.offset >> PAGE_SHIFT);
1552}
1553#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1554
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555/*
David Howells8feae132009-01-08 12:04:47 +00001556 * split a vma into two pieces at address 'addr', a new vma is allocated either
1557 * for the first part or the tail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 */
David Howells8feae132009-01-08 12:04:47 +00001559int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1560 unsigned long addr, int new_below)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561{
David Howells8feae132009-01-08 12:04:47 +00001562 struct vm_area_struct *new;
1563 struct vm_region *region;
1564 unsigned long npages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565
David Howells8feae132009-01-08 12:04:47 +00001566 kenter("");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567
David Howells779c1022010-01-15 17:01:34 -08001568 /* we're only permitted to split anonymous regions (these should have
1569 * only a single usage on the region) */
1570 if (vma->vm_file)
David Howells8feae132009-01-08 12:04:47 +00001571 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572
David Howells8feae132009-01-08 12:04:47 +00001573 if (mm->map_count >= sysctl_max_map_count)
1574 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575
David Howells8feae132009-01-08 12:04:47 +00001576 region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1577 if (!region)
1578 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579
David Howells8feae132009-01-08 12:04:47 +00001580 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1581 if (!new) {
1582 kmem_cache_free(vm_region_jar, region);
1583 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584 }
David Howells8feae132009-01-08 12:04:47 +00001585
1586 /* most fields are the same, copy all, and then fixup */
1587 *new = *vma;
1588 *region = *vma->vm_region;
1589 new->vm_region = region;
1590
1591 npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1592
1593 if (new_below) {
Paul Mundtdd8632a2009-01-08 12:04:47 +00001594 region->vm_top = region->vm_end = new->vm_end = addr;
David Howells8feae132009-01-08 12:04:47 +00001595 } else {
1596 region->vm_start = new->vm_start = addr;
1597 region->vm_pgoff = new->vm_pgoff += npages;
1598 }
1599
1600 if (new->vm_ops && new->vm_ops->open)
1601 new->vm_ops->open(new);
1602
1603 delete_vma_from_mm(vma);
1604 down_write(&nommu_region_sem);
1605 delete_nommu_region(vma->vm_region);
1606 if (new_below) {
1607 vma->vm_region->vm_start = vma->vm_start = addr;
1608 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1609 } else {
1610 vma->vm_region->vm_end = vma->vm_end = addr;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001611 vma->vm_region->vm_top = addr;
David Howells8feae132009-01-08 12:04:47 +00001612 }
1613 add_nommu_region(vma->vm_region);
1614 add_nommu_region(new->vm_region);
1615 up_write(&nommu_region_sem);
1616 add_vma_to_mm(mm, vma);
1617 add_vma_to_mm(mm, new);
1618 return 0;
1619}
1620
1621/*
1622 * shrink a VMA by removing the specified chunk from either the beginning or
1623 * the end
1624 */
1625static int shrink_vma(struct mm_struct *mm,
1626 struct vm_area_struct *vma,
1627 unsigned long from, unsigned long to)
1628{
1629 struct vm_region *region;
1630
1631 kenter("");
1632
1633 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1634 * and list */
1635 delete_vma_from_mm(vma);
1636 if (from > vma->vm_start)
1637 vma->vm_end = from;
1638 else
1639 vma->vm_start = to;
1640 add_vma_to_mm(mm, vma);
1641
1642 /* cut the backing region down to size */
1643 region = vma->vm_region;
David Howells1e2ae592010-01-15 17:01:33 -08001644 BUG_ON(region->vm_usage != 1);
David Howells8feae132009-01-08 12:04:47 +00001645
1646 down_write(&nommu_region_sem);
1647 delete_nommu_region(region);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001648 if (from > region->vm_start) {
1649 to = region->vm_top;
1650 region->vm_top = region->vm_end = from;
1651 } else {
David Howells8feae132009-01-08 12:04:47 +00001652 region->vm_start = to;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001653 }
David Howells8feae132009-01-08 12:04:47 +00001654 add_nommu_region(region);
1655 up_write(&nommu_region_sem);
1656
1657 free_page_series(from, to);
1658 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659}
1660
David Howells30340972006-09-27 01:50:20 -07001661/*
1662 * release a mapping
David Howells8feae132009-01-08 12:04:47 +00001663 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1664 * VMA, though it need not cover the whole VMA
David Howells30340972006-09-27 01:50:20 -07001665 */
David Howells8feae132009-01-08 12:04:47 +00001666int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667{
David Howells8feae132009-01-08 12:04:47 +00001668 struct vm_area_struct *vma;
Bob Liuf67d9b12011-05-24 17:12:56 -07001669 unsigned long end;
David Howells8feae132009-01-08 12:04:47 +00001670 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671
David Howells8feae132009-01-08 12:04:47 +00001672 kenter(",%lx,%zx", start, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673
Bob Liuf67d9b12011-05-24 17:12:56 -07001674 len = PAGE_ALIGN(len);
David Howells8feae132009-01-08 12:04:47 +00001675 if (len == 0)
1676 return -EINVAL;
1677
Bob Liuf67d9b12011-05-24 17:12:56 -07001678 end = start + len;
1679
David Howells8feae132009-01-08 12:04:47 +00001680 /* find the first potentially overlapping VMA */
1681 vma = find_vma(mm, start);
1682 if (!vma) {
Choi Gi-yongac714902014-04-07 15:37:36 -07001683 static int limit;
David Howells33e5d7692009-04-02 16:56:32 -07001684 if (limit < 5) {
1685 printk(KERN_WARNING
1686 "munmap of memory not mmapped by process %d"
1687 " (%s): 0x%lx-0x%lx\n",
1688 current->pid, current->comm,
1689 start, start + len - 1);
1690 limit++;
1691 }
David Howells8feae132009-01-08 12:04:47 +00001692 return -EINVAL;
David Howells30340972006-09-27 01:50:20 -07001693 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694
David Howells8feae132009-01-08 12:04:47 +00001695 /* we're allowed to split an anonymous VMA but not a file-backed one */
1696 if (vma->vm_file) {
1697 do {
1698 if (start > vma->vm_start) {
1699 kleave(" = -EINVAL [miss]");
1700 return -EINVAL;
1701 }
1702 if (end == vma->vm_end)
1703 goto erase_whole_vma;
Namhyung Kimd75a3102011-05-24 17:11:25 -07001704 vma = vma->vm_next;
1705 } while (vma);
David Howells8feae132009-01-08 12:04:47 +00001706 kleave(" = -EINVAL [split file]");
1707 return -EINVAL;
1708 } else {
1709 /* the chunk must be a subset of the VMA found */
1710 if (start == vma->vm_start && end == vma->vm_end)
1711 goto erase_whole_vma;
1712 if (start < vma->vm_start || end > vma->vm_end) {
1713 kleave(" = -EINVAL [superset]");
1714 return -EINVAL;
1715 }
1716 if (start & ~PAGE_MASK) {
1717 kleave(" = -EINVAL [unaligned start]");
1718 return -EINVAL;
1719 }
1720 if (end != vma->vm_end && end & ~PAGE_MASK) {
1721 kleave(" = -EINVAL [unaligned split]");
1722 return -EINVAL;
1723 }
1724 if (start != vma->vm_start && end != vma->vm_end) {
1725 ret = split_vma(mm, vma, start, 1);
1726 if (ret < 0) {
1727 kleave(" = %d [split]", ret);
1728 return ret;
1729 }
1730 }
1731 return shrink_vma(mm, vma, start, end);
1732 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733
David Howells8feae132009-01-08 12:04:47 +00001734erase_whole_vma:
1735 delete_vma_from_mm(vma);
1736 delete_vma(mm, vma);
1737 kleave(" = 0");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738 return 0;
1739}
Paul Mundtb5073172007-07-21 04:37:25 -07001740EXPORT_SYMBOL(do_munmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741
Al Virobfce2812012-04-20 21:57:04 -04001742int vm_munmap(unsigned long addr, size_t len)
David Howells30340972006-09-27 01:50:20 -07001743{
Al Virobfce2812012-04-20 21:57:04 -04001744 struct mm_struct *mm = current->mm;
David Howells30340972006-09-27 01:50:20 -07001745 int ret;
David Howells30340972006-09-27 01:50:20 -07001746
1747 down_write(&mm->mmap_sem);
1748 ret = do_munmap(mm, addr, len);
1749 up_write(&mm->mmap_sem);
1750 return ret;
1751}
Linus Torvaldsa46ef992012-04-20 16:20:01 -07001752EXPORT_SYMBOL(vm_munmap);
1753
1754SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1755{
Al Virobfce2812012-04-20 21:57:04 -04001756 return vm_munmap(addr, len);
Linus Torvaldsa46ef992012-04-20 16:20:01 -07001757}
David Howells30340972006-09-27 01:50:20 -07001758
1759/*
David Howells8feae132009-01-08 12:04:47 +00001760 * release all the mappings made in a process's VM space
David Howells30340972006-09-27 01:50:20 -07001761 */
David Howells8feae132009-01-08 12:04:47 +00001762void exit_mmap(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763{
David Howells8feae132009-01-08 12:04:47 +00001764 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765
David Howells8feae132009-01-08 12:04:47 +00001766 if (!mm)
1767 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768
David Howells8feae132009-01-08 12:04:47 +00001769 kenter("");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770
David Howells8feae132009-01-08 12:04:47 +00001771 mm->total_vm = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772
David Howells8feae132009-01-08 12:04:47 +00001773 while ((vma = mm->mmap)) {
1774 mm->mmap = vma->vm_next;
1775 delete_vma_from_mm(vma);
1776 delete_vma(mm, vma);
Steven J. Magnani04c34962010-11-24 12:56:54 -08001777 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778 }
David Howells8feae132009-01-08 12:04:47 +00001779
1780 kleave("");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781}
1782
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07001783unsigned long vm_brk(unsigned long addr, unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784{
1785 return -ENOMEM;
1786}
1787
1788/*
David Howells6fa5f802006-09-27 01:50:21 -07001789 * expand (or shrink) an existing mapping, potentially moving it at the same
1790 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791 *
David Howells6fa5f802006-09-27 01:50:21 -07001792 * under NOMMU conditions, we only permit changing a mapping's size, and only
David Howells8feae132009-01-08 12:04:47 +00001793 * as long as it stays within the region allocated by do_mmap_private() and the
1794 * block is not shareable
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795 *
David Howells6fa5f802006-09-27 01:50:21 -07001796 * MREMAP_FIXED is not supported under NOMMU conditions
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797 */
Al Viro4b377ba2013-03-04 10:47:59 -05001798static unsigned long do_mremap(unsigned long addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799 unsigned long old_len, unsigned long new_len,
1800 unsigned long flags, unsigned long new_addr)
1801{
David Howells6fa5f802006-09-27 01:50:21 -07001802 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803
1804 /* insanity checks first */
Bob Liuf67d9b12011-05-24 17:12:56 -07001805 old_len = PAGE_ALIGN(old_len);
1806 new_len = PAGE_ALIGN(new_len);
David Howells8feae132009-01-08 12:04:47 +00001807 if (old_len == 0 || new_len == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808 return (unsigned long) -EINVAL;
1809
David Howells8feae132009-01-08 12:04:47 +00001810 if (addr & ~PAGE_MASK)
1811 return -EINVAL;
1812
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813 if (flags & MREMAP_FIXED && new_addr != addr)
1814 return (unsigned long) -EINVAL;
1815
David Howells8feae132009-01-08 12:04:47 +00001816 vma = find_vma_exact(current->mm, addr, old_len);
David Howells6fa5f802006-09-27 01:50:21 -07001817 if (!vma)
1818 return (unsigned long) -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819
David Howells6fa5f802006-09-27 01:50:21 -07001820 if (vma->vm_end != vma->vm_start + old_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821 return (unsigned long) -EFAULT;
1822
David Howells6fa5f802006-09-27 01:50:21 -07001823 if (vma->vm_flags & VM_MAYSHARE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824 return (unsigned long) -EPERM;
1825
David Howells8feae132009-01-08 12:04:47 +00001826 if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827 return (unsigned long) -ENOMEM;
1828
1829 /* all checks complete - do it */
David Howells6fa5f802006-09-27 01:50:21 -07001830 vma->vm_end = vma->vm_start + new_len;
David Howells6fa5f802006-09-27 01:50:21 -07001831 return vma->vm_start;
1832}
1833
Heiko Carstens6a6160a2009-01-14 14:14:15 +01001834SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1835 unsigned long, new_len, unsigned long, flags,
1836 unsigned long, new_addr)
David Howells6fa5f802006-09-27 01:50:21 -07001837{
1838 unsigned long ret;
1839
1840 down_write(&current->mm->mmap_sem);
1841 ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1842 up_write(&current->mm->mmap_sem);
1843 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844}
1845
Michel Lespinasse240aade2013-02-22 16:35:56 -08001846struct page *follow_page_mask(struct vm_area_struct *vma,
1847 unsigned long address, unsigned int flags,
1848 unsigned int *page_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849{
Michel Lespinasse240aade2013-02-22 16:35:56 -08001850 *page_mask = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851 return NULL;
1852}
1853
Bob Liu8f3b1322011-07-08 15:39:46 -07001854int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1855 unsigned long pfn, unsigned long size, pgprot_t prot)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856{
Bob Liu8f3b1322011-07-08 15:39:46 -07001857 if (addr != (pfn << PAGE_SHIFT))
1858 return -EINVAL;
1859
Konstantin Khlebnikov314e51b2012-10-08 16:29:02 -07001860 vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
Greg Ungerer66aa2b42005-09-12 11:18:10 +10001861 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862}
Luke Yang22c4af42006-07-14 00:24:09 -07001863EXPORT_SYMBOL(remap_pfn_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864
Linus Torvalds3c0b9de2013-04-27 13:25:38 -07001865int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)
1866{
1867 unsigned long pfn = start >> PAGE_SHIFT;
1868 unsigned long vm_len = vma->vm_end - vma->vm_start;
1869
1870 pfn += vma->vm_pgoff;
1871 return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);
1872}
1873EXPORT_SYMBOL(vm_iomap_memory);
1874
Paul Mundtf905bc42008-02-04 22:29:59 -08001875int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1876 unsigned long pgoff)
1877{
1878 unsigned int size = vma->vm_end - vma->vm_start;
1879
1880 if (!(vma->vm_flags & VM_USERMAP))
1881 return -EINVAL;
1882
1883 vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1884 vma->vm_end = vma->vm_start + size;
1885
1886 return 0;
1887}
1888EXPORT_SYMBOL(remap_vmalloc_range);
1889
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1891 unsigned long len, unsigned long pgoff, unsigned long flags)
1892{
1893 return -ENOMEM;
1894}
1895
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896void unmap_mapping_range(struct address_space *mapping,
1897 loff_t const holebegin, loff_t const holelen,
1898 int even_cows)
1899{
1900}
Luke Yang22c4af42006-07-14 00:24:09 -07001901EXPORT_SYMBOL(unmap_mapping_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902
1903/*
1904 * Check that a process has enough memory to allocate a new virtual
1905 * mapping. 0 means there is enough memory for the allocation to
1906 * succeed and -ENOMEM implies there is not.
1907 *
1908 * We currently support three overcommit policies, which are set via the
1909 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
1910 *
1911 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1912 * Additional code 2002 Jul 20 by Robert Love.
1913 *
1914 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1915 *
1916 * Note this is a helper function intended to be used by LSMs which
1917 * wish to use this logic.
1918 */
Alan Cox34b4e4a2007-08-22 14:01:28 -07001919int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920{
Roman Gushchin8138a67a2015-02-11 15:28:42 -08001921 long free, allowed, reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922
1923 vm_acct_memory(pages);
1924
1925 /*
1926 * Sometimes we want to use more memory than we have
1927 */
1928 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1929 return 0;
1930
1931 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
Dmitry Finkc15bef32011-07-25 17:12:19 -07001932 free = global_page_state(NR_FREE_PAGES);
1933 free += global_page_state(NR_FILE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934
Dmitry Finkc15bef32011-07-25 17:12:19 -07001935 /*
1936 * shmem pages shouldn't be counted as free in this
1937 * case, they can't be purged, only swapped out, and
1938 * that won't affect the overall amount of available
1939 * memory in the system.
1940 */
1941 free -= global_page_state(NR_SHMEM);
1942
Shaohua Liec8acf22013-02-22 16:34:38 -08001943 free += get_nr_swap_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944
1945 /*
1946 * Any slabs which are created with the
1947 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1948 * which are reclaimable, under pressure. The dentry
1949 * cache and most inode caches should fall into this
1950 */
Christoph Lameter972d1a72006-09-25 23:31:51 -07001951 free += global_page_state(NR_SLAB_RECLAIMABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952
1953 /*
Dmitry Finkc15bef32011-07-25 17:12:19 -07001954 * Leave reserved pages. The pages are not for anonymous pages.
1955 */
1956 if (free <= totalreserve_pages)
1957 goto error;
1958 else
1959 free -= totalreserve_pages;
1960
1961 /*
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07001962 * Reserve some for root
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963 */
1964 if (!cap_sys_admin)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07001965 free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966
1967 if (free > pages)
1968 return 0;
1969
Hideo AOKId5ddc792006-04-10 22:53:01 -07001970 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971 }
1972
Jerome Marchand00619bc2013-11-12 15:08:31 -08001973 allowed = vm_commit_limit();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974 /*
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07001975 * Reserve some 3% for root
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976 */
1977 if (!cap_sys_admin)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07001978 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07001980 /*
1981 * Don't let a single process grow so big a user can't recover
1982 */
1983 if (mm) {
1984 reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
Roman Gushchin8138a67a2015-02-11 15:28:42 -08001985 allowed -= min_t(long, mm->total_vm / 32, reserve);
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07001986 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07001988 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989 return 0;
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07001990
Hideo AOKId5ddc792006-04-10 22:53:01 -07001991error:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992 vm_unacct_memory(pages);
1993
1994 return -ENOMEM;
1995}
1996
Nick Piggind0217ac2007-07-19 01:47:03 -07001997int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
David Howellsb0e15192006-01-06 00:11:42 -08001998{
1999 BUG();
Nick Piggind0217ac2007-07-19 01:47:03 -07002000 return 0;
David Howellsb0e15192006-01-06 00:11:42 -08002001}
Paul Mundtb5073172007-07-21 04:37:25 -07002002EXPORT_SYMBOL(filemap_fault);
David Howells0ec76a12006-09-27 01:50:15 -07002003
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002004void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf)
2005{
2006 BUG();
2007}
2008EXPORT_SYMBOL(filemap_map_pages);
2009
Mike Frysingerf55f1992011-03-29 14:05:12 +01002010static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
2011 unsigned long addr, void *buf, int len, int write)
David Howells0ec76a12006-09-27 01:50:15 -07002012{
David Howells0ec76a12006-09-27 01:50:15 -07002013 struct vm_area_struct *vma;
David Howells0ec76a12006-09-27 01:50:15 -07002014
2015 down_read(&mm->mmap_sem);
2016
2017 /* the access must start within one of the target process's mappings */
David Howells0159b142006-09-27 01:50:16 -07002018 vma = find_vma(mm, addr);
2019 if (vma) {
David Howells0ec76a12006-09-27 01:50:15 -07002020 /* don't overrun this mapping */
2021 if (addr + len >= vma->vm_end)
2022 len = vma->vm_end - addr;
2023
2024 /* only read or write mappings where it is permitted */
David Howellsd00c7b992006-09-27 01:50:19 -07002025 if (write && vma->vm_flags & VM_MAYWRITE)
Jie Zhang79597222010-01-06 17:23:28 +00002026 copy_to_user_page(vma, NULL, addr,
2027 (void *) addr, buf, len);
David Howellsd00c7b992006-09-27 01:50:19 -07002028 else if (!write && vma->vm_flags & VM_MAYREAD)
Jie Zhang79597222010-01-06 17:23:28 +00002029 copy_from_user_page(vma, NULL, addr,
2030 buf, (void *) addr, len);
David Howells0ec76a12006-09-27 01:50:15 -07002031 else
2032 len = 0;
2033 } else {
2034 len = 0;
2035 }
2036
2037 up_read(&mm->mmap_sem);
Mike Frysingerf55f1992011-03-29 14:05:12 +01002038
2039 return len;
2040}
2041
2042/**
2043 * @access_remote_vm - access another process' address space
2044 * @mm: the mm_struct of the target address space
2045 * @addr: start address to access
2046 * @buf: source or destination buffer
2047 * @len: number of bytes to transfer
2048 * @write: whether the access is a write
2049 *
2050 * The caller must hold a reference on @mm.
2051 */
2052int access_remote_vm(struct mm_struct *mm, unsigned long addr,
2053 void *buf, int len, int write)
2054{
2055 return __access_remote_vm(NULL, mm, addr, buf, len, write);
2056}
2057
2058/*
2059 * Access another process' address space.
2060 * - source/target buffer must be kernel space
2061 */
2062int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
2063{
2064 struct mm_struct *mm;
2065
2066 if (addr + len < addr)
2067 return 0;
2068
2069 mm = get_task_mm(tsk);
2070 if (!mm)
2071 return 0;
2072
2073 len = __access_remote_vm(tsk, mm, addr, buf, len, write);
2074
David Howells0ec76a12006-09-27 01:50:15 -07002075 mmput(mm);
2076 return len;
2077}
David Howells7e660872010-01-15 17:01:39 -08002078
2079/**
2080 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
2081 * @inode: The inode to check
2082 * @size: The current filesize of the inode
2083 * @newsize: The proposed filesize of the inode
2084 *
2085 * Check the shared mappings on an inode on behalf of a shrinking truncate to
2086 * make sure that that any outstanding VMAs aren't broken and then shrink the
2087 * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
2088 * automatically grant mappings that are too large.
2089 */
2090int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
2091 size_t newsize)
2092{
2093 struct vm_area_struct *vma;
David Howells7e660872010-01-15 17:01:39 -08002094 struct vm_region *region;
2095 pgoff_t low, high;
2096 size_t r_size, r_top;
2097
2098 low = newsize >> PAGE_SHIFT;
2099 high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
2100
2101 down_write(&nommu_region_sem);
Davidlohr Bueso1acf2e02014-12-12 16:54:39 -08002102 i_mmap_lock_read(inode->i_mapping);
David Howells7e660872010-01-15 17:01:39 -08002103
2104 /* search for VMAs that fall within the dead zone */
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -07002105 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) {
David Howells7e660872010-01-15 17:01:39 -08002106 /* found one - only interested if it's shared out of the page
2107 * cache */
2108 if (vma->vm_flags & VM_SHARED) {
Davidlohr Bueso1acf2e02014-12-12 16:54:39 -08002109 i_mmap_unlock_read(inode->i_mapping);
David Howells7e660872010-01-15 17:01:39 -08002110 up_write(&nommu_region_sem);
2111 return -ETXTBSY; /* not quite true, but near enough */
2112 }
2113 }
2114
2115 /* reduce any regions that overlap the dead zone - if in existence,
2116 * these will be pointed to by VMAs that don't overlap the dead zone
2117 *
2118 * we don't check for any regions that start beyond the EOF as there
2119 * shouldn't be any
2120 */
Davidlohr Bueso1acf2e02014-12-12 16:54:39 -08002121 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, 0, ULONG_MAX) {
David Howells7e660872010-01-15 17:01:39 -08002122 if (!(vma->vm_flags & VM_SHARED))
2123 continue;
2124
2125 region = vma->vm_region;
2126 r_size = region->vm_top - region->vm_start;
2127 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
2128
2129 if (r_top > newsize) {
2130 region->vm_top -= r_top - newsize;
2131 if (region->vm_end > region->vm_top)
2132 region->vm_end = region->vm_top;
2133 }
2134 }
2135
Davidlohr Bueso1acf2e02014-12-12 16:54:39 -08002136 i_mmap_unlock_read(inode->i_mapping);
David Howells7e660872010-01-15 17:01:39 -08002137 up_write(&nommu_region_sem);
2138 return 0;
2139}
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07002140
2141/*
2142 * Initialise sysctl_user_reserve_kbytes.
2143 *
2144 * This is intended to prevent a user from starting a single memory hogging
2145 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
2146 * mode.
2147 *
2148 * The default value is min(3% of free memory, 128MB)
2149 * 128MB is enough to recover with sshd/login, bash, and top/kill.
2150 */
2151static int __meminit init_user_reserve(void)
2152{
2153 unsigned long free_kbytes;
2154
2155 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
2156
2157 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
2158 return 0;
2159}
2160module_init(init_user_reserve)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07002161
2162/*
2163 * Initialise sysctl_admin_reserve_kbytes.
2164 *
2165 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
2166 * to log in and kill a memory hogging process.
2167 *
2168 * Systems with more than 256MB will reserve 8MB, enough to recover
2169 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
2170 * only reserve 3% of free pages by default.
2171 */
2172static int __meminit init_admin_reserve(void)
2173{
2174 unsigned long free_kbytes;
2175
2176 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
2177
2178 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
2179 return 0;
2180}
2181module_init(init_admin_reserve)