blob: a881d9673c6b165e5336e83bf2ab75bf3c03d938 [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;
62struct page *mem_map;
63unsigned long max_mapnr;
Hugh Dickins4266c972009-09-23 17:05:53 +010064unsigned long highest_memmap_pfn;
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -070065struct percpu_counter vm_committed_as;
Linus Torvalds1da177e2005-04-16 15:20:36 -070066int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
67int sysctl_overcommit_ratio = 50; /* default is 50% */
Jerome Marchand49f0ce52014-01-21 15:49:14 -080068unsigned long sysctl_overcommit_kbytes __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070069int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
David Howellsfc4d5c22009-05-06 16:03:05 -070070int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -070071unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -070072unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
Linus Torvalds1da177e2005-04-16 15:20:36 -070073int heap_stack_gap = 0;
74
David Howells33e5d7692009-04-02 16:56:32 -070075atomic_long_t mmap_pages_allocated;
David Howells8feae132009-01-08 12:04:47 +000076
K. Y. Srinivasan997071b2012-11-15 14:34:42 -080077/*
78 * The global memory commitment made in the system can be a metric
79 * that can be used to drive ballooning decisions when Linux is hosted
80 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
81 * balancing memory across competing virtual machines that are hosted.
82 * Several metrics drive this policy engine including the guest reported
83 * memory commitment.
84 */
85unsigned long vm_memory_committed(void)
86{
87 return percpu_counter_read_positive(&vm_committed_as);
88}
89
90EXPORT_SYMBOL_GPL(vm_memory_committed);
91
Linus Torvalds1da177e2005-04-16 15:20:36 -070092EXPORT_SYMBOL(mem_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -070093
David Howells8feae132009-01-08 12:04:47 +000094/* list of mapped, potentially shareable regions */
95static struct kmem_cache *vm_region_jar;
96struct rb_root nommu_region_tree = RB_ROOT;
97DECLARE_RWSEM(nommu_region_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -070098
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +040099const struct vm_operations_struct generic_file_vm_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100};
101
102/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103 * Return the total memory allocated for this pointer, not
104 * just what the caller asked for.
105 *
106 * Doesn't have to be accurate, i.e. may have races.
107 */
108unsigned int kobjsize(const void *objp)
109{
110 struct page *page;
111
Michael Hennerich4016a132008-04-28 02:13:38 -0700112 /*
113 * If the object we have should not have ksize performed on it,
114 * return size of 0
115 */
Paul Mundt5a1603b2008-06-12 16:29:55 +0900116 if (!objp || !virt_addr_valid(objp))
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700117 return 0;
118
119 page = virt_to_head_page(objp);
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700120
121 /*
122 * If the allocator sets PageSlab, we know the pointer came from
123 * kmalloc().
124 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125 if (PageSlab(page))
126 return ksize(objp);
127
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700128 /*
Paul Mundtab2e83e2009-01-08 12:04:48 +0000129 * If it's not a compound page, see if we have a matching VMA
130 * region. This test is intentionally done in reverse order,
131 * so if there's no VMA, we still fall through and hand back
132 * PAGE_SIZE for 0-order pages.
133 */
134 if (!PageCompound(page)) {
135 struct vm_area_struct *vma;
136
137 vma = find_vma(current->mm, (unsigned long)objp);
138 if (vma)
139 return vma->vm_end - vma->vm_start;
140 }
141
142 /*
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700143 * The ksize() function is only guaranteed to work for pointers
Paul Mundt5a1603b2008-06-12 16:29:55 +0900144 * returned by kmalloc(). So handle arbitrary pointers here.
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700145 */
Paul Mundt5a1603b2008-06-12 16:29:55 +0900146 return PAGE_SIZE << compound_order(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147}
148
Michel Lespinasse28a35712013-02-22 16:35:55 -0800149long __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
150 unsigned long start, unsigned long nr_pages,
151 unsigned int foll_flags, struct page **pages,
152 struct vm_area_struct **vmas, int *nonblocking)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153{
Sonic Zhang910e46d2006-09-27 01:50:17 -0700154 struct vm_area_struct *vma;
David Howells7b4d5b82006-09-27 01:50:18 -0700155 unsigned long vm_flags;
156 int i;
157
158 /* calculate required read or write permissions.
Hugh Dickins58fa8792009-09-21 17:03:31 -0700159 * If FOLL_FORCE is set, we only require the "MAY" flags.
David Howells7b4d5b82006-09-27 01:50:18 -0700160 */
Hugh Dickins58fa8792009-09-21 17:03:31 -0700161 vm_flags = (foll_flags & FOLL_WRITE) ?
162 (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
163 vm_flags &= (foll_flags & FOLL_FORCE) ?
164 (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165
Peter Zijlstra9d737772009-06-25 11:58:55 +0200166 for (i = 0; i < nr_pages; i++) {
David Howells7561e8c2010-03-25 16:48:38 +0000167 vma = find_vma(mm, start);
David Howells7b4d5b82006-09-27 01:50:18 -0700168 if (!vma)
169 goto finish_or_fault;
170
171 /* protect what we can, including chardevs */
Hugh Dickins1c3aff12009-09-21 17:03:24 -0700172 if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
173 !(vm_flags & vma->vm_flags))
David Howells7b4d5b82006-09-27 01:50:18 -0700174 goto finish_or_fault;
Sonic Zhang910e46d2006-09-27 01:50:17 -0700175
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 if (pages) {
177 pages[i] = virt_to_page(start);
178 if (pages[i])
179 page_cache_get(pages[i]);
180 }
181 if (vmas)
Sonic Zhang910e46d2006-09-27 01:50:17 -0700182 vmas[i] = vma;
David Howellse1ee65d2010-03-25 16:48:44 +0000183 start = (start + PAGE_SIZE) & PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184 }
David Howells7b4d5b82006-09-27 01:50:18 -0700185
186 return i;
187
188finish_or_fault:
189 return i ? : -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190}
Nick Pigginb291f002008-10-18 20:26:44 -0700191
Nick Pigginb291f002008-10-18 20:26:44 -0700192/*
193 * get a list of pages in an address range belonging to the specified process
194 * and indicate the VMA that covers each page
195 * - this is potentially dodgy as we may end incrementing the page count of a
196 * slab page or a secondary page from a compound page
197 * - don't permit access to VMAs that don't support it, such as I/O mappings
198 */
Michel Lespinasse28a35712013-02-22 16:35:55 -0800199long get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
200 unsigned long start, unsigned long nr_pages,
201 int write, int force, struct page **pages,
202 struct vm_area_struct **vmas)
Nick Pigginb291f002008-10-18 20:26:44 -0700203{
204 int flags = 0;
205
206 if (write)
Hugh Dickins58fa8792009-09-21 17:03:31 -0700207 flags |= FOLL_WRITE;
Nick Pigginb291f002008-10-18 20:26:44 -0700208 if (force)
Hugh Dickins58fa8792009-09-21 17:03:31 -0700209 flags |= FOLL_FORCE;
Nick Pigginb291f002008-10-18 20:26:44 -0700210
Michel Lespinasse53a77062011-01-13 15:46:14 -0800211 return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas,
212 NULL);
Nick Pigginb291f002008-10-18 20:26:44 -0700213}
Greg Ungerer66aa2b42005-09-12 11:18:10 +1000214EXPORT_SYMBOL(get_user_pages);
215
Paul Mundtdfc2f912009-06-26 04:31:57 +0900216/**
217 * follow_pfn - look up PFN at a user virtual address
218 * @vma: memory mapping
219 * @address: user virtual address
220 * @pfn: location to store found PFN
221 *
222 * Only IO mappings and raw PFN mappings are allowed.
223 *
224 * Returns zero and the pfn at @pfn on success, -ve otherwise.
225 */
226int follow_pfn(struct vm_area_struct *vma, unsigned long address,
227 unsigned long *pfn)
228{
229 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
230 return -EINVAL;
231
232 *pfn = address >> PAGE_SHIFT;
233 return 0;
234}
235EXPORT_SYMBOL(follow_pfn);
236
Joonsoo Kimf1c40692013-04-29 15:07:37 -0700237LIST_HEAD(vmap_area_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800239void vfree(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240{
241 kfree(addr);
242}
Paul Mundtb5073172007-07-21 04:37:25 -0700243EXPORT_SYMBOL(vfree);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244
Al Virodd0fc662005-10-07 07:46:04 +0100245void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246{
247 /*
Robert P. J. Day85186092007-10-19 23:11:38 +0200248 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
249 * returns only a logical address.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 */
Nick Piggin84097512006-03-22 00:08:34 -0800251 return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252}
Paul Mundtb5073172007-07-21 04:37:25 -0700253EXPORT_SYMBOL(__vmalloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254
Paul Mundtf905bc42008-02-04 22:29:59 -0800255void *vmalloc_user(unsigned long size)
256{
257 void *ret;
258
259 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
260 PAGE_KERNEL);
261 if (ret) {
262 struct vm_area_struct *vma;
263
264 down_write(&current->mm->mmap_sem);
265 vma = find_vma(current->mm, (unsigned long)ret);
266 if (vma)
267 vma->vm_flags |= VM_USERMAP;
268 up_write(&current->mm->mmap_sem);
269 }
270
271 return ret;
272}
273EXPORT_SYMBOL(vmalloc_user);
274
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800275struct page *vmalloc_to_page(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276{
277 return virt_to_page(addr);
278}
Paul Mundtb5073172007-07-21 04:37:25 -0700279EXPORT_SYMBOL(vmalloc_to_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800281unsigned long vmalloc_to_pfn(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282{
283 return page_to_pfn(virt_to_page(addr));
284}
Paul Mundtb5073172007-07-21 04:37:25 -0700285EXPORT_SYMBOL(vmalloc_to_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286
287long vread(char *buf, char *addr, unsigned long count)
288{
Chen Gang9bde9162013-07-03 15:02:36 -0700289 /* Don't allow overflow */
290 if ((unsigned long) buf + count < count)
291 count = -(unsigned long) buf;
292
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293 memcpy(buf, addr, count);
294 return count;
295}
296
297long vwrite(char *buf, char *addr, unsigned long count)
298{
299 /* Don't allow overflow */
300 if ((unsigned long) addr + count < count)
301 count = -(unsigned long) addr;
302
303 memcpy(addr, buf, count);
Choi Gi-yongac714902014-04-07 15:37:36 -0700304 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305}
306
307/*
308 * vmalloc - allocate virtually continguos memory
309 *
310 * @size: allocation size
311 *
312 * Allocate enough pages to cover @size from the page level
313 * allocator and map them into continguos kernel virtual space.
314 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +0200315 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 * use __vmalloc() instead.
317 */
318void *vmalloc(unsigned long size)
319{
320 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
321}
Andrew Mortonf6138882006-02-28 16:59:18 -0800322EXPORT_SYMBOL(vmalloc);
323
Dave Younge1ca7782010-10-26 14:22:06 -0700324/*
325 * vzalloc - allocate virtually continguos memory with zero fill
326 *
327 * @size: allocation size
328 *
329 * Allocate enough pages to cover @size from the page level
330 * allocator and map them into continguos kernel virtual space.
331 * The memory allocated is set to zero.
332 *
333 * For tight control over page level allocator and protection flags
334 * use __vmalloc() instead.
335 */
336void *vzalloc(unsigned long size)
337{
338 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
339 PAGE_KERNEL);
340}
341EXPORT_SYMBOL(vzalloc);
342
343/**
344 * vmalloc_node - allocate memory on a specific node
345 * @size: allocation size
346 * @node: numa node
347 *
348 * Allocate enough pages to cover @size from the page level
349 * allocator and map them into contiguous kernel virtual space.
350 *
351 * For tight control over page level allocator and protection flags
352 * use __vmalloc() instead.
353 */
Andrew Mortonf6138882006-02-28 16:59:18 -0800354void *vmalloc_node(unsigned long size, int node)
355{
356 return vmalloc(size);
357}
Paul Mundt9a14f652010-12-24 11:50:34 +0900358EXPORT_SYMBOL(vmalloc_node);
Dave Younge1ca7782010-10-26 14:22:06 -0700359
360/**
361 * vzalloc_node - allocate memory on a specific node with zero fill
362 * @size: allocation size
363 * @node: numa node
364 *
365 * Allocate enough pages to cover @size from the page level
366 * allocator and map them into contiguous kernel virtual space.
367 * The memory allocated is set to zero.
368 *
369 * For tight control over page level allocator and protection flags
370 * use __vmalloc() instead.
371 */
372void *vzalloc_node(unsigned long size, int node)
373{
374 return vzalloc(size);
375}
376EXPORT_SYMBOL(vzalloc_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377
Paul Mundt1af446e2008-08-04 16:01:47 +0900378#ifndef PAGE_KERNEL_EXEC
379# define PAGE_KERNEL_EXEC PAGE_KERNEL
380#endif
381
382/**
383 * vmalloc_exec - allocate virtually contiguous, executable memory
384 * @size: allocation size
385 *
386 * Kernel-internal function to allocate enough pages to cover @size
387 * the page level allocator and map them into contiguous and
388 * executable kernel virtual space.
389 *
390 * For tight control over page level allocator and protection flags
391 * use __vmalloc() instead.
392 */
393
394void *vmalloc_exec(unsigned long size)
395{
396 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
397}
398
Paul Mundtb5073172007-07-21 04:37:25 -0700399/**
400 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401 * @size: allocation size
402 *
403 * Allocate enough 32bit PA addressable pages to cover @size from the
404 * page level allocator and map them into continguos kernel virtual space.
405 */
406void *vmalloc_32(unsigned long size)
407{
408 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
409}
Paul Mundtb5073172007-07-21 04:37:25 -0700410EXPORT_SYMBOL(vmalloc_32);
411
412/**
413 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
414 * @size: allocation size
415 *
416 * The resulting memory area is 32bit addressable and zeroed so it can be
417 * mapped to userspace without leaking data.
Paul Mundtf905bc42008-02-04 22:29:59 -0800418 *
419 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
420 * remap_vmalloc_range() are permissible.
Paul Mundtb5073172007-07-21 04:37:25 -0700421 */
422void *vmalloc_32_user(unsigned long size)
423{
Paul Mundtf905bc42008-02-04 22:29:59 -0800424 /*
425 * We'll have to sort out the ZONE_DMA bits for 64-bit,
426 * but for now this can simply use vmalloc_user() directly.
427 */
428 return vmalloc_user(size);
Paul Mundtb5073172007-07-21 04:37:25 -0700429}
430EXPORT_SYMBOL(vmalloc_32_user);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431
432void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
433{
434 BUG();
435 return NULL;
436}
Paul Mundtb5073172007-07-21 04:37:25 -0700437EXPORT_SYMBOL(vmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800439void vunmap(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440{
441 BUG();
442}
Paul Mundtb5073172007-07-21 04:37:25 -0700443EXPORT_SYMBOL(vunmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444
Paul Mundteb6434d2009-01-21 17:45:47 +0900445void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
446{
447 BUG();
448 return NULL;
449}
450EXPORT_SYMBOL(vm_map_ram);
451
452void vm_unmap_ram(const void *mem, unsigned int count)
453{
454 BUG();
455}
456EXPORT_SYMBOL(vm_unmap_ram);
457
458void vm_unmap_aliases(void)
459{
460}
461EXPORT_SYMBOL_GPL(vm_unmap_aliases);
462
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463/*
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -0700464 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
465 * have one.
466 */
Gideon Israel Dsouza3b321232014-04-07 15:37:26 -0700467void __weak vmalloc_sync_all(void)
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -0700468{
469}
470
Paul Mundt29c185e2010-12-24 12:08:30 +0900471/**
472 * alloc_vm_area - allocate a range of kernel address space
473 * @size: size of the area
474 *
475 * Returns: NULL on failure, vm_struct on success
476 *
477 * This function reserves a range of kernel address space, and
478 * allocates pagetables to map that range. No actual mappings
479 * are created. If the kernel address space is not shared
480 * between processes, it syncs the pagetable across all
481 * processes.
482 */
David Vrabelcd129092011-09-29 16:53:32 +0100483struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
Paul Mundt29c185e2010-12-24 12:08:30 +0900484{
485 BUG();
486 return NULL;
487}
488EXPORT_SYMBOL_GPL(alloc_vm_area);
489
490void free_vm_area(struct vm_struct *area)
491{
492 BUG();
493}
494EXPORT_SYMBOL_GPL(free_vm_area);
495
Paul Mundtb5073172007-07-21 04:37:25 -0700496int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
497 struct page *page)
498{
499 return -EINVAL;
500}
501EXPORT_SYMBOL(vm_insert_page);
502
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -0700503/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 * sys_brk() for the most part doesn't need the global kernel
505 * lock, except when an application is doing something nasty
506 * like trying to un-brk an area that has already been mapped
507 * to a regular file. in this case, the unmapping will need
508 * to invoke file system routines that need the global lock.
509 */
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100510SYSCALL_DEFINE1(brk, unsigned long, brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511{
512 struct mm_struct *mm = current->mm;
513
514 if (brk < mm->start_brk || brk > mm->context.end_brk)
515 return mm->brk;
516
517 if (mm->brk == brk)
518 return mm->brk;
519
520 /*
521 * Always allow shrinking brk
522 */
523 if (brk <= mm->brk) {
524 mm->brk = brk;
525 return brk;
526 }
527
528 /*
529 * Ok, looks good - let it rip.
530 */
Mike Frysingercfe79c02010-01-06 17:23:23 +0000531 flush_icache_range(mm->brk, brk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 return mm->brk = brk;
533}
534
David Howells8feae132009-01-08 12:04:47 +0000535/*
536 * initialise the VMA and region record slabs
537 */
538void __init mmap_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539{
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -0700540 int ret;
541
542 ret = percpu_counter_init(&vm_committed_as, 0);
543 VM_BUG_ON(ret);
David Howells33e5d7692009-04-02 16:56:32 -0700544 vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC);
David Howells8feae132009-01-08 12:04:47 +0000545}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546
David Howells8feae132009-01-08 12:04:47 +0000547/*
548 * validate the region tree
549 * - the caller must hold the region lock
550 */
551#ifdef CONFIG_DEBUG_NOMMU_REGIONS
552static noinline void validate_nommu_regions(void)
553{
554 struct vm_region *region, *last;
555 struct rb_node *p, *lastp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556
David Howells8feae132009-01-08 12:04:47 +0000557 lastp = rb_first(&nommu_region_tree);
558 if (!lastp)
559 return;
560
561 last = rb_entry(lastp, struct vm_region, vm_rb);
David Howells33e5d7692009-04-02 16:56:32 -0700562 BUG_ON(unlikely(last->vm_end <= last->vm_start));
563 BUG_ON(unlikely(last->vm_top < last->vm_end));
David Howells8feae132009-01-08 12:04:47 +0000564
565 while ((p = rb_next(lastp))) {
566 region = rb_entry(p, struct vm_region, vm_rb);
567 last = rb_entry(lastp, struct vm_region, vm_rb);
568
David Howells33e5d7692009-04-02 16:56:32 -0700569 BUG_ON(unlikely(region->vm_end <= region->vm_start));
570 BUG_ON(unlikely(region->vm_top < region->vm_end));
571 BUG_ON(unlikely(region->vm_start < last->vm_top));
David Howells8feae132009-01-08 12:04:47 +0000572
573 lastp = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 }
575}
David Howells8feae132009-01-08 12:04:47 +0000576#else
David Howells33e5d7692009-04-02 16:56:32 -0700577static void validate_nommu_regions(void)
578{
579}
David Howells8feae132009-01-08 12:04:47 +0000580#endif
581
582/*
583 * add a region into the global tree
584 */
585static void add_nommu_region(struct vm_region *region)
586{
587 struct vm_region *pregion;
588 struct rb_node **p, *parent;
589
590 validate_nommu_regions();
591
David Howells8feae132009-01-08 12:04:47 +0000592 parent = NULL;
593 p = &nommu_region_tree.rb_node;
594 while (*p) {
595 parent = *p;
596 pregion = rb_entry(parent, struct vm_region, vm_rb);
597 if (region->vm_start < pregion->vm_start)
598 p = &(*p)->rb_left;
599 else if (region->vm_start > pregion->vm_start)
600 p = &(*p)->rb_right;
601 else if (pregion == region)
602 return;
603 else
604 BUG();
605 }
606
607 rb_link_node(&region->vm_rb, parent, p);
608 rb_insert_color(&region->vm_rb, &nommu_region_tree);
609
610 validate_nommu_regions();
611}
612
613/*
614 * delete a region from the global tree
615 */
616static void delete_nommu_region(struct vm_region *region)
617{
618 BUG_ON(!nommu_region_tree.rb_node);
619
620 validate_nommu_regions();
621 rb_erase(&region->vm_rb, &nommu_region_tree);
622 validate_nommu_regions();
623}
624
625/*
626 * free a contiguous series of pages
627 */
628static void free_page_series(unsigned long from, unsigned long to)
629{
630 for (; from < to; from += PAGE_SIZE) {
631 struct page *page = virt_to_page(from);
632
633 kdebug("- free %lx", from);
David Howells33e5d7692009-04-02 16:56:32 -0700634 atomic_long_dec(&mmap_pages_allocated);
David Howells8feae132009-01-08 12:04:47 +0000635 if (page_count(page) != 1)
David Howells33e5d7692009-04-02 16:56:32 -0700636 kdebug("free page %p: refcount not one: %d",
637 page, page_count(page));
David Howells8feae132009-01-08 12:04:47 +0000638 put_page(page);
639 }
640}
641
642/*
643 * release a reference to a region
David Howells33e5d7692009-04-02 16:56:32 -0700644 * - the caller must hold the region semaphore for writing, which this releases
Paul Mundtdd8632a2009-01-08 12:04:47 +0000645 * - the region may not have been added to the tree yet, in which case vm_top
David Howells8feae132009-01-08 12:04:47 +0000646 * will equal vm_start
647 */
648static void __put_nommu_region(struct vm_region *region)
649 __releases(nommu_region_sem)
650{
David Howells1e2ae592010-01-15 17:01:33 -0800651 kenter("%p{%d}", region, region->vm_usage);
David Howells8feae132009-01-08 12:04:47 +0000652
653 BUG_ON(!nommu_region_tree.rb_node);
654
David Howells1e2ae592010-01-15 17:01:33 -0800655 if (--region->vm_usage == 0) {
Paul Mundtdd8632a2009-01-08 12:04:47 +0000656 if (region->vm_top > region->vm_start)
David Howells8feae132009-01-08 12:04:47 +0000657 delete_nommu_region(region);
658 up_write(&nommu_region_sem);
659
660 if (region->vm_file)
661 fput(region->vm_file);
662
663 /* IO memory and memory shared directly out of the pagecache
664 * from ramfs/tmpfs mustn't be released here */
665 if (region->vm_flags & VM_MAPPED_COPY) {
666 kdebug("free series");
Paul Mundtdd8632a2009-01-08 12:04:47 +0000667 free_page_series(region->vm_start, region->vm_top);
David Howells8feae132009-01-08 12:04:47 +0000668 }
669 kmem_cache_free(vm_region_jar, region);
670 } else {
671 up_write(&nommu_region_sem);
672 }
673}
674
675/*
676 * release a reference to a region
677 */
678static void put_nommu_region(struct vm_region *region)
679{
680 down_write(&nommu_region_sem);
681 __put_nommu_region(region);
682}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683
David Howells30340972006-09-27 01:50:20 -0700684/*
Bernd Schmidteb8cdec2009-09-21 17:03:57 -0700685 * update protection on a vma
686 */
687static void protect_vma(struct vm_area_struct *vma, unsigned long flags)
688{
689#ifdef CONFIG_MPU
690 struct mm_struct *mm = vma->vm_mm;
691 long start = vma->vm_start & PAGE_MASK;
692 while (start < vma->vm_end) {
693 protect_page(mm, start, flags);
694 start += PAGE_SIZE;
695 }
696 update_protections(mm);
697#endif
698}
699
700/*
David Howells30340972006-09-27 01:50:20 -0700701 * add a VMA into a process's mm_struct in the appropriate place in the list
David Howells8feae132009-01-08 12:04:47 +0000702 * and tree and add to the address space's page tree also if not an anonymous
703 * page
David Howells30340972006-09-27 01:50:20 -0700704 * - should be called with mm->mmap_sem held writelocked
705 */
David Howells8feae132009-01-08 12:04:47 +0000706static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
David Howells30340972006-09-27 01:50:20 -0700707{
Namhyung Kim6038def2011-05-24 17:11:22 -0700708 struct vm_area_struct *pvma, *prev;
David Howells8feae132009-01-08 12:04:47 +0000709 struct address_space *mapping;
Namhyung Kim6038def2011-05-24 17:11:22 -0700710 struct rb_node **p, *parent, *rb_prev;
David Howells30340972006-09-27 01:50:20 -0700711
David Howells8feae132009-01-08 12:04:47 +0000712 kenter(",%p", vma);
713
714 BUG_ON(!vma->vm_region);
715
716 mm->map_count++;
717 vma->vm_mm = mm;
718
Bernd Schmidteb8cdec2009-09-21 17:03:57 -0700719 protect_vma(vma, vma->vm_flags);
720
David Howells8feae132009-01-08 12:04:47 +0000721 /* add the VMA to the mapping */
722 if (vma->vm_file) {
723 mapping = vma->vm_file->f_mapping;
724
David Howells918e5562012-02-23 13:50:35 +0000725 mutex_lock(&mapping->i_mmap_mutex);
David Howells8feae132009-01-08 12:04:47 +0000726 flush_dcache_mmap_lock(mapping);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700727 vma_interval_tree_insert(vma, &mapping->i_mmap);
David Howells8feae132009-01-08 12:04:47 +0000728 flush_dcache_mmap_unlock(mapping);
David Howells918e5562012-02-23 13:50:35 +0000729 mutex_unlock(&mapping->i_mmap_mutex);
David Howells8feae132009-01-08 12:04:47 +0000730 }
731
732 /* add the VMA to the tree */
Namhyung Kim6038def2011-05-24 17:11:22 -0700733 parent = rb_prev = NULL;
David Howells8feae132009-01-08 12:04:47 +0000734 p = &mm->mm_rb.rb_node;
735 while (*p) {
736 parent = *p;
737 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
738
739 /* sort by: start addr, end addr, VMA struct addr in that order
740 * (the latter is necessary as we may get identical VMAs) */
741 if (vma->vm_start < pvma->vm_start)
742 p = &(*p)->rb_left;
Namhyung Kim6038def2011-05-24 17:11:22 -0700743 else if (vma->vm_start > pvma->vm_start) {
744 rb_prev = parent;
David Howells8feae132009-01-08 12:04:47 +0000745 p = &(*p)->rb_right;
Namhyung Kim6038def2011-05-24 17:11:22 -0700746 } else if (vma->vm_end < pvma->vm_end)
David Howells8feae132009-01-08 12:04:47 +0000747 p = &(*p)->rb_left;
Namhyung Kim6038def2011-05-24 17:11:22 -0700748 else if (vma->vm_end > pvma->vm_end) {
749 rb_prev = parent;
David Howells8feae132009-01-08 12:04:47 +0000750 p = &(*p)->rb_right;
Namhyung Kim6038def2011-05-24 17:11:22 -0700751 } else if (vma < pvma)
David Howells8feae132009-01-08 12:04:47 +0000752 p = &(*p)->rb_left;
Namhyung Kim6038def2011-05-24 17:11:22 -0700753 else if (vma > pvma) {
754 rb_prev = parent;
David Howells8feae132009-01-08 12:04:47 +0000755 p = &(*p)->rb_right;
Namhyung Kim6038def2011-05-24 17:11:22 -0700756 } else
David Howells8feae132009-01-08 12:04:47 +0000757 BUG();
758 }
759
760 rb_link_node(&vma->vm_rb, parent, p);
761 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
762
763 /* add VMA to the VMA list also */
Namhyung Kim6038def2011-05-24 17:11:22 -0700764 prev = NULL;
765 if (rb_prev)
766 prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
David Howells30340972006-09-27 01:50:20 -0700767
Namhyung Kim6038def2011-05-24 17:11:22 -0700768 __vma_link_list(mm, vma, prev, parent);
David Howells8feae132009-01-08 12:04:47 +0000769}
770
771/*
772 * delete a VMA from its owning mm_struct and address space
773 */
774static void delete_vma_from_mm(struct vm_area_struct *vma)
775{
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700776 int i;
David Howells8feae132009-01-08 12:04:47 +0000777 struct address_space *mapping;
778 struct mm_struct *mm = vma->vm_mm;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700779 struct task_struct *curr = current;
David Howells8feae132009-01-08 12:04:47 +0000780
781 kenter("%p", vma);
782
Bernd Schmidteb8cdec2009-09-21 17:03:57 -0700783 protect_vma(vma, 0);
784
David Howells8feae132009-01-08 12:04:47 +0000785 mm->map_count--;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700786 for (i = 0; i < VMACACHE_SIZE; i++) {
787 /* if the vma is cached, invalidate the entire cache */
788 if (curr->vmacache[i] == vma) {
Steven Miaoe020d5b2014-06-23 13:22:02 -0700789 vmacache_invalidate(mm);
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700790 break;
791 }
792 }
David Howells8feae132009-01-08 12:04:47 +0000793
794 /* remove the VMA from the mapping */
795 if (vma->vm_file) {
796 mapping = vma->vm_file->f_mapping;
797
David Howells918e5562012-02-23 13:50:35 +0000798 mutex_lock(&mapping->i_mmap_mutex);
David Howells8feae132009-01-08 12:04:47 +0000799 flush_dcache_mmap_lock(mapping);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700800 vma_interval_tree_remove(vma, &mapping->i_mmap);
David Howells8feae132009-01-08 12:04:47 +0000801 flush_dcache_mmap_unlock(mapping);
David Howells918e5562012-02-23 13:50:35 +0000802 mutex_unlock(&mapping->i_mmap_mutex);
David Howells8feae132009-01-08 12:04:47 +0000803 }
804
805 /* remove from the MM's tree and list */
806 rb_erase(&vma->vm_rb, &mm->mm_rb);
Namhyung Kimb951bf22011-05-24 17:11:23 -0700807
808 if (vma->vm_prev)
809 vma->vm_prev->vm_next = vma->vm_next;
810 else
811 mm->mmap = vma->vm_next;
812
813 if (vma->vm_next)
814 vma->vm_next->vm_prev = vma->vm_prev;
David Howells8feae132009-01-08 12:04:47 +0000815}
816
817/*
818 * destroy a VMA record
819 */
820static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
821{
822 kenter("%p", vma);
823 if (vma->vm_ops && vma->vm_ops->close)
824 vma->vm_ops->close(vma);
Konstantin Khlebnikove9714ac2012-10-08 16:28:54 -0700825 if (vma->vm_file)
David Howells8feae132009-01-08 12:04:47 +0000826 fput(vma->vm_file);
David Howells8feae132009-01-08 12:04:47 +0000827 put_nommu_region(vma->vm_region);
828 kmem_cache_free(vm_area_cachep, vma);
David Howells30340972006-09-27 01:50:20 -0700829}
830
831/*
832 * look up the first VMA in which addr resides, NULL if none
833 * - should be called with mm->mmap_sem at least held readlocked
834 */
835struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
836{
David Howells8feae132009-01-08 12:04:47 +0000837 struct vm_area_struct *vma;
David Howells30340972006-09-27 01:50:20 -0700838
David Howells8feae132009-01-08 12:04:47 +0000839 /* check the cache first */
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700840 vma = vmacache_find(mm, addr);
841 if (likely(vma))
David Howells8feae132009-01-08 12:04:47 +0000842 return vma;
843
Namhyung Kime922c4c2011-05-24 17:11:24 -0700844 /* trawl the list (there may be multiple mappings in which addr
David Howells8feae132009-01-08 12:04:47 +0000845 * resides) */
Namhyung Kime922c4c2011-05-24 17:11:24 -0700846 for (vma = mm->mmap; vma; vma = vma->vm_next) {
David Howells8feae132009-01-08 12:04:47 +0000847 if (vma->vm_start > addr)
848 return NULL;
849 if (vma->vm_end > addr) {
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700850 vmacache_update(addr, vma);
David Howells8feae132009-01-08 12:04:47 +0000851 return vma;
852 }
David Howells30340972006-09-27 01:50:20 -0700853 }
854
David Howells30340972006-09-27 01:50:20 -0700855 return NULL;
856}
857EXPORT_SYMBOL(find_vma);
858
859/*
David Howells930e6522006-09-27 01:50:22 -0700860 * find a VMA
861 * - we don't extend stack VMAs under NOMMU conditions
862 */
863struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
864{
David Howells7561e8c2010-03-25 16:48:38 +0000865 return find_vma(mm, addr);
David Howells930e6522006-09-27 01:50:22 -0700866}
867
David Howells8feae132009-01-08 12:04:47 +0000868/*
869 * expand a stack to a given address
870 * - not supported under NOMMU conditions
871 */
Greg Ungerer57c8f632007-07-15 23:38:28 -0700872int expand_stack(struct vm_area_struct *vma, unsigned long address)
873{
874 return -ENOMEM;
875}
876
David Howells930e6522006-09-27 01:50:22 -0700877/*
David Howells6fa5f802006-09-27 01:50:21 -0700878 * look up the first VMA exactly that exactly matches addr
879 * - should be called with mm->mmap_sem at least held readlocked
880 */
David Howells8feae132009-01-08 12:04:47 +0000881static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
882 unsigned long addr,
883 unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884{
885 struct vm_area_struct *vma;
David Howells8feae132009-01-08 12:04:47 +0000886 unsigned long end = addr + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887
David Howells8feae132009-01-08 12:04:47 +0000888 /* check the cache first */
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700889 vma = vmacache_find_exact(mm, addr, end);
890 if (vma)
David Howells8feae132009-01-08 12:04:47 +0000891 return vma;
892
Namhyung Kime922c4c2011-05-24 17:11:24 -0700893 /* trawl the list (there may be multiple mappings in which addr
David Howells8feae132009-01-08 12:04:47 +0000894 * resides) */
Namhyung Kime922c4c2011-05-24 17:11:24 -0700895 for (vma = mm->mmap; vma; vma = vma->vm_next) {
David Howells8feae132009-01-08 12:04:47 +0000896 if (vma->vm_start < addr)
897 continue;
898 if (vma->vm_start > addr)
899 return NULL;
900 if (vma->vm_end == end) {
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700901 vmacache_update(addr, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 return vma;
David Howells8feae132009-01-08 12:04:47 +0000903 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904 }
905
906 return NULL;
907}
908
David Howells30340972006-09-27 01:50:20 -0700909/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 * determine whether a mapping should be permitted and, if so, what sort of
911 * mapping we're capable of supporting
912 */
913static int validate_mmap_request(struct file *file,
914 unsigned long addr,
915 unsigned long len,
916 unsigned long prot,
917 unsigned long flags,
918 unsigned long pgoff,
919 unsigned long *_capabilities)
920{
David Howells8feae132009-01-08 12:04:47 +0000921 unsigned long capabilities, rlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 int ret;
923
924 /* do the simple checks first */
David Howells06aab5a2009-09-24 12:33:48 +0100925 if (flags & MAP_FIXED) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926 printk(KERN_DEBUG
927 "%d: Can't do fixed-address/overlay mmap of RAM\n",
928 current->pid);
929 return -EINVAL;
930 }
931
932 if ((flags & MAP_TYPE) != MAP_PRIVATE &&
933 (flags & MAP_TYPE) != MAP_SHARED)
934 return -EINVAL;
935
Mike Frysingerf81cff02006-12-06 12:02:59 +1000936 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 return -EINVAL;
938
Mike Frysingerf81cff02006-12-06 12:02:59 +1000939 /* Careful about overflows.. */
David Howells8feae132009-01-08 12:04:47 +0000940 rlen = PAGE_ALIGN(len);
941 if (!rlen || rlen > TASK_SIZE)
Mike Frysingerf81cff02006-12-06 12:02:59 +1000942 return -ENOMEM;
943
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 /* offset overflow? */
David Howells8feae132009-01-08 12:04:47 +0000945 if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
Mike Frysingerf81cff02006-12-06 12:02:59 +1000946 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947
948 if (file) {
949 /* validate file mapping requests */
950 struct address_space *mapping;
951
952 /* files must support mmap */
Al Viro72c2d532013-09-22 16:27:52 -0400953 if (!file->f_op->mmap)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954 return -ENODEV;
955
956 /* work out if what we've got could possibly be shared
957 * - we support chardevs that provide their own "memory"
958 * - we support files/blockdevs that are memory backed
959 */
960 mapping = file->f_mapping;
961 if (!mapping)
Al Viro496ad9a2013-01-23 17:07:38 -0500962 mapping = file_inode(file)->i_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963
964 capabilities = 0;
965 if (mapping && mapping->backing_dev_info)
966 capabilities = mapping->backing_dev_info->capabilities;
967
968 if (!capabilities) {
969 /* no explicit capabilities set, so assume some
970 * defaults */
Al Viro496ad9a2013-01-23 17:07:38 -0500971 switch (file_inode(file)->i_mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 case S_IFREG:
973 case S_IFBLK:
974 capabilities = BDI_CAP_MAP_COPY;
975 break;
976
977 case S_IFCHR:
978 capabilities =
979 BDI_CAP_MAP_DIRECT |
980 BDI_CAP_READ_MAP |
981 BDI_CAP_WRITE_MAP;
982 break;
983
984 default:
985 return -EINVAL;
986 }
987 }
988
989 /* eliminate any capabilities that we can't support on this
990 * device */
991 if (!file->f_op->get_unmapped_area)
992 capabilities &= ~BDI_CAP_MAP_DIRECT;
993 if (!file->f_op->read)
994 capabilities &= ~BDI_CAP_MAP_COPY;
995
Graff Yang28d7a6a2009-08-18 14:11:17 -0700996 /* The file shall have been opened with read permission. */
997 if (!(file->f_mode & FMODE_READ))
998 return -EACCES;
999
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 if (flags & MAP_SHARED) {
1001 /* do checks for writing, appending and locking */
1002 if ((prot & PROT_WRITE) &&
1003 !(file->f_mode & FMODE_WRITE))
1004 return -EACCES;
1005
Al Viro496ad9a2013-01-23 17:07:38 -05001006 if (IS_APPEND(file_inode(file)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 (file->f_mode & FMODE_WRITE))
1008 return -EACCES;
1009
Jeff Laytond7a06982014-03-10 09:54:15 -04001010 if (locks_verify_locked(file))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 return -EAGAIN;
1012
1013 if (!(capabilities & BDI_CAP_MAP_DIRECT))
1014 return -ENODEV;
1015
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016 /* we mustn't privatise shared mappings */
1017 capabilities &= ~BDI_CAP_MAP_COPY;
Choi Gi-yongac714902014-04-07 15:37:36 -07001018 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 /* we're going to read the file into private memory we
1020 * allocate */
1021 if (!(capabilities & BDI_CAP_MAP_COPY))
1022 return -ENODEV;
1023
1024 /* we don't permit a private writable mapping to be
1025 * shared with the backing device */
1026 if (prot & PROT_WRITE)
1027 capabilities &= ~BDI_CAP_MAP_DIRECT;
1028 }
1029
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001030 if (capabilities & BDI_CAP_MAP_DIRECT) {
1031 if (((prot & PROT_READ) && !(capabilities & BDI_CAP_READ_MAP)) ||
1032 ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
1033 ((prot & PROT_EXEC) && !(capabilities & BDI_CAP_EXEC_MAP))
1034 ) {
1035 capabilities &= ~BDI_CAP_MAP_DIRECT;
1036 if (flags & MAP_SHARED) {
1037 printk(KERN_WARNING
1038 "MAP_SHARED not completely supported on !MMU\n");
1039 return -EINVAL;
1040 }
1041 }
1042 }
1043
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 /* handle executable mappings and implied executable
1045 * mappings */
Josef Sipeke9536ae2006-12-08 02:37:21 -08001046 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 if (prot & PROT_EXEC)
1048 return -EPERM;
Choi Gi-yongac714902014-04-07 15:37:36 -07001049 } else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 /* handle implication of PROT_EXEC by PROT_READ */
1051 if (current->personality & READ_IMPLIES_EXEC) {
1052 if (capabilities & BDI_CAP_EXEC_MAP)
1053 prot |= PROT_EXEC;
1054 }
Choi Gi-yongac714902014-04-07 15:37:36 -07001055 } else if ((prot & PROT_READ) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 (prot & PROT_EXEC) &&
1057 !(capabilities & BDI_CAP_EXEC_MAP)
1058 ) {
1059 /* backing file is not executable, try to copy */
1060 capabilities &= ~BDI_CAP_MAP_DIRECT;
1061 }
Choi Gi-yongac714902014-04-07 15:37:36 -07001062 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 /* anonymous mappings are always memory backed and can be
1064 * privately mapped
1065 */
1066 capabilities = BDI_CAP_MAP_COPY;
1067
1068 /* handle PROT_EXEC implication by PROT_READ */
1069 if ((prot & PROT_READ) &&
1070 (current->personality & READ_IMPLIES_EXEC))
1071 prot |= PROT_EXEC;
1072 }
1073
1074 /* allow the security API to have its say */
Al Viroe5467852012-05-30 13:30:51 -04001075 ret = security_mmap_addr(addr);
1076 if (ret < 0)
1077 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078
1079 /* looks okay */
1080 *_capabilities = capabilities;
1081 return 0;
1082}
1083
1084/*
1085 * we've determined that we can make the mapping, now translate what we
1086 * now know into VMA flags
1087 */
1088static unsigned long determine_vm_flags(struct file *file,
1089 unsigned long prot,
1090 unsigned long flags,
1091 unsigned long capabilities)
1092{
1093 unsigned long vm_flags;
1094
1095 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 /* vm_flags |= mm->def_flags; */
1097
1098 if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
1099 /* attempt to share read-only copies of mapped file chunks */
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001100 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 if (file && !(prot & PROT_WRITE))
1102 vm_flags |= VM_MAYSHARE;
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001103 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104 /* overlay a shareable mapping on the backing device or inode
1105 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1106 * romfs/cramfs */
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001107 vm_flags |= VM_MAYSHARE | (capabilities & BDI_CAP_VMFLAGS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 if (flags & MAP_SHARED)
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001109 vm_flags |= VM_SHARED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 }
1111
1112 /* refuse to let anyone share private mappings with this process if
1113 * it's being traced - otherwise breakpoints set in it may interfere
1114 * with another untraced process
1115 */
Tejun Heoa288eec2011-06-17 16:50:37 +02001116 if ((flags & MAP_PRIVATE) && current->ptrace)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 vm_flags &= ~VM_MAYSHARE;
1118
1119 return vm_flags;
1120}
1121
1122/*
David Howells8feae132009-01-08 12:04:47 +00001123 * set up a shared mapping on a file (the driver or filesystem provides and
1124 * pins the storage)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 */
David Howells8feae132009-01-08 12:04:47 +00001126static int do_mmap_shared_file(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127{
1128 int ret;
1129
1130 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001131 if (ret == 0) {
1132 vma->vm_region->vm_top = vma->vm_region->vm_end;
David Howells645d83c2009-09-24 15:13:10 +01001133 return 0;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001134 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 if (ret != -ENOSYS)
1136 return ret;
1137
David Howells3fa30462010-03-23 13:35:21 -07001138 /* getting -ENOSYS indicates that direct mmap isn't possible (as
1139 * opposed to tried but failed) so we can only give a suitable error as
1140 * it's not possible to make a private copy if MAP_SHARED was given */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141 return -ENODEV;
1142}
1143
1144/*
1145 * set up a private mapping or an anonymous shared mapping
1146 */
David Howells8feae132009-01-08 12:04:47 +00001147static int do_mmap_private(struct vm_area_struct *vma,
1148 struct vm_region *region,
David Howells645d83c2009-09-24 15:13:10 +01001149 unsigned long len,
1150 unsigned long capabilities)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151{
David Howells8feae132009-01-08 12:04:47 +00001152 struct page *pages;
Bob Liuf67d9b12011-05-24 17:12:56 -07001153 unsigned long total, point, n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 void *base;
David Howells8feae132009-01-08 12:04:47 +00001155 int ret, order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156
1157 /* invoke the file's mapping function so that it can keep track of
1158 * shared mappings on devices or memory
1159 * - VM_MAYSHARE will be set if it may attempt to share
1160 */
David Howells645d83c2009-09-24 15:13:10 +01001161 if (capabilities & BDI_CAP_MAP_DIRECT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001163 if (ret == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 /* shouldn't return success if we're not sharing */
Paul Mundtdd8632a2009-01-08 12:04:47 +00001165 BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1166 vma->vm_region->vm_top = vma->vm_region->vm_end;
David Howells645d83c2009-09-24 15:13:10 +01001167 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 }
Paul Mundtdd8632a2009-01-08 12:04:47 +00001169 if (ret != -ENOSYS)
1170 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171
1172 /* getting an ENOSYS error indicates that direct mmap isn't
1173 * possible (as opposed to tried but failed) so we'll try to
1174 * make a private copy of the data and map that instead */
1175 }
1176
David Howells8feae132009-01-08 12:04:47 +00001177
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 /* allocate some memory to hold the mapping
1179 * - note that this may not return a page-aligned address if the object
1180 * we're allocating is smaller than a page
1181 */
Bob Liuf67d9b12011-05-24 17:12:56 -07001182 order = get_order(len);
David Howells8feae132009-01-08 12:04:47 +00001183 kdebug("alloc order %d for %lx", order, len);
1184
1185 pages = alloc_pages(GFP_KERNEL, order);
1186 if (!pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187 goto enomem;
1188
David Howells8feae132009-01-08 12:04:47 +00001189 total = 1 << order;
David Howells33e5d7692009-04-02 16:56:32 -07001190 atomic_long_add(total, &mmap_pages_allocated);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191
Bob Liuf67d9b12011-05-24 17:12:56 -07001192 point = len >> PAGE_SHIFT;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001193
1194 /* we allocated a power-of-2 sized page set, so we may want to trim off
1195 * the excess */
1196 if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) {
1197 while (total > point) {
1198 order = ilog2(total - point);
1199 n = 1 << order;
1200 kdebug("shave %lu/%lu @%lu", n, total - point, total);
David Howells33e5d7692009-04-02 16:56:32 -07001201 atomic_long_sub(n, &mmap_pages_allocated);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001202 total -= n;
1203 set_page_refcounted(pages + total);
1204 __free_pages(pages + total, order);
1205 }
David Howells8feae132009-01-08 12:04:47 +00001206 }
1207
David Howells8feae132009-01-08 12:04:47 +00001208 for (point = 1; point < total; point++)
1209 set_page_refcounted(&pages[point]);
1210
1211 base = page_address(pages);
1212 region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1213 region->vm_start = (unsigned long) base;
Bob Liuf67d9b12011-05-24 17:12:56 -07001214 region->vm_end = region->vm_start + len;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001215 region->vm_top = region->vm_start + (total << PAGE_SHIFT);
David Howells8feae132009-01-08 12:04:47 +00001216
1217 vma->vm_start = region->vm_start;
1218 vma->vm_end = region->vm_start + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219
1220 if (vma->vm_file) {
1221 /* read the contents of a file into the copy */
1222 mm_segment_t old_fs;
1223 loff_t fpos;
1224
1225 fpos = vma->vm_pgoff;
1226 fpos <<= PAGE_SHIFT;
1227
1228 old_fs = get_fs();
1229 set_fs(KERNEL_DS);
Bob Liuf67d9b12011-05-24 17:12:56 -07001230 ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 set_fs(old_fs);
1232
1233 if (ret < 0)
1234 goto error_free;
1235
1236 /* clear the last little bit */
Bob Liuf67d9b12011-05-24 17:12:56 -07001237 if (ret < len)
1238 memset(base + ret, 0, len - ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 }
1241
1242 return 0;
1243
1244error_free:
Namhyung Kim7223bb42011-05-24 17:11:26 -07001245 free_page_series(region->vm_start, region->vm_top);
David Howells8feae132009-01-08 12:04:47 +00001246 region->vm_start = vma->vm_start = 0;
1247 region->vm_end = vma->vm_end = 0;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001248 region->vm_top = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249 return ret;
1250
1251enomem:
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -07001252 pr_err("Allocation of length %lu from process %d (%s) failed\n",
Greg Ungerer05ae6fa2009-01-13 17:30:22 +10001253 len, current->pid, current->comm);
David Rientjes7bf02ea2011-05-24 17:11:16 -07001254 show_free_areas(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 return -ENOMEM;
1256}
1257
1258/*
1259 * handle mapping creation for uClinux
1260 */
Al Viroe3fc6292012-05-30 20:08:42 -04001261unsigned long do_mmap_pgoff(struct file *file,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 unsigned long addr,
1263 unsigned long len,
1264 unsigned long prot,
1265 unsigned long flags,
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001266 unsigned long pgoff,
Michel Lespinasse41badc12013-02-22 16:32:47 -08001267 unsigned long *populate)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268{
David Howells8feae132009-01-08 12:04:47 +00001269 struct vm_area_struct *vma;
1270 struct vm_region *region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 struct rb_node *rb;
David Howells8feae132009-01-08 12:04:47 +00001272 unsigned long capabilities, vm_flags, result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 int ret;
1274
David Howells8feae132009-01-08 12:04:47 +00001275 kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff);
1276
Michel Lespinasse41badc12013-02-22 16:32:47 -08001277 *populate = 0;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001278
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 /* decide whether we should attempt the mapping, and if so what sort of
1280 * mapping */
1281 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1282 &capabilities);
David Howells8feae132009-01-08 12:04:47 +00001283 if (ret < 0) {
1284 kleave(" = %d [val]", ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 return ret;
David Howells8feae132009-01-08 12:04:47 +00001286 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287
David Howells06aab5a2009-09-24 12:33:48 +01001288 /* we ignore the address hint */
1289 addr = 0;
Bob Liuf67d9b12011-05-24 17:12:56 -07001290 len = PAGE_ALIGN(len);
David Howells06aab5a2009-09-24 12:33:48 +01001291
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292 /* we've determined that we can make the mapping, now translate what we
1293 * now know into VMA flags */
1294 vm_flags = determine_vm_flags(file, prot, flags, capabilities);
1295
David Howells8feae132009-01-08 12:04:47 +00001296 /* we're going to need to record the mapping */
1297 region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1298 if (!region)
1299 goto error_getting_region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300
David Howells8feae132009-01-08 12:04:47 +00001301 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1302 if (!vma)
1303 goto error_getting_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304
David Howells1e2ae592010-01-15 17:01:33 -08001305 region->vm_usage = 1;
David Howells8feae132009-01-08 12:04:47 +00001306 region->vm_flags = vm_flags;
1307 region->vm_pgoff = pgoff;
1308
Rik van Riel5beb4932010-03-05 13:42:07 -08001309 INIT_LIST_HEAD(&vma->anon_vma_chain);
David Howells8feae132009-01-08 12:04:47 +00001310 vma->vm_flags = vm_flags;
1311 vma->vm_pgoff = pgoff;
1312
1313 if (file) {
Al Virocb0942b2012-08-27 14:48:26 -04001314 region->vm_file = get_file(file);
1315 vma->vm_file = get_file(file);
David Howells8feae132009-01-08 12:04:47 +00001316 }
1317
1318 down_write(&nommu_region_sem);
1319
1320 /* if we want to share, we need to check for regions created by other
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321 * mmap() calls that overlap with our proposed mapping
David Howells8feae132009-01-08 12:04:47 +00001322 * - we can only share with a superset match on most regular files
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323 * - shared mappings on character devices and memory backed files are
1324 * permitted to overlap inexactly as far as we are concerned for in
1325 * these cases, sharing is handled in the driver or filesystem rather
1326 * than here
1327 */
1328 if (vm_flags & VM_MAYSHARE) {
David Howells8feae132009-01-08 12:04:47 +00001329 struct vm_region *pregion;
1330 unsigned long pglen, rpglen, pgend, rpgend, start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331
David Howells8feae132009-01-08 12:04:47 +00001332 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1333 pgend = pgoff + pglen;
David Howells165b2392007-03-22 00:11:24 -08001334
David Howells8feae132009-01-08 12:04:47 +00001335 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1336 pregion = rb_entry(rb, struct vm_region, vm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337
David Howells8feae132009-01-08 12:04:47 +00001338 if (!(pregion->vm_flags & VM_MAYSHARE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 continue;
1340
1341 /* search for overlapping mappings on the same file */
Al Viro496ad9a2013-01-23 17:07:38 -05001342 if (file_inode(pregion->vm_file) !=
1343 file_inode(file))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 continue;
1345
David Howells8feae132009-01-08 12:04:47 +00001346 if (pregion->vm_pgoff >= pgend)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347 continue;
1348
David Howells8feae132009-01-08 12:04:47 +00001349 rpglen = pregion->vm_end - pregion->vm_start;
1350 rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1351 rpgend = pregion->vm_pgoff + rpglen;
1352 if (pgoff >= rpgend)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353 continue;
1354
David Howells8feae132009-01-08 12:04:47 +00001355 /* handle inexactly overlapping matches between
1356 * mappings */
1357 if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1358 !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1359 /* new mapping is not a subset of the region */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360 if (!(capabilities & BDI_CAP_MAP_DIRECT))
1361 goto sharing_violation;
1362 continue;
1363 }
1364
David Howells8feae132009-01-08 12:04:47 +00001365 /* we've found a region we can share */
David Howells1e2ae592010-01-15 17:01:33 -08001366 pregion->vm_usage++;
David Howells8feae132009-01-08 12:04:47 +00001367 vma->vm_region = pregion;
1368 start = pregion->vm_start;
1369 start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1370 vma->vm_start = start;
1371 vma->vm_end = start + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372
David Howells8feae132009-01-08 12:04:47 +00001373 if (pregion->vm_flags & VM_MAPPED_COPY) {
1374 kdebug("share copy");
1375 vma->vm_flags |= VM_MAPPED_COPY;
1376 } else {
1377 kdebug("share mmap");
1378 ret = do_mmap_shared_file(vma);
1379 if (ret < 0) {
1380 vma->vm_region = NULL;
1381 vma->vm_start = 0;
1382 vma->vm_end = 0;
David Howells1e2ae592010-01-15 17:01:33 -08001383 pregion->vm_usage--;
David Howells8feae132009-01-08 12:04:47 +00001384 pregion = NULL;
1385 goto error_just_free;
1386 }
1387 }
1388 fput(region->vm_file);
1389 kmem_cache_free(vm_region_jar, region);
1390 region = pregion;
1391 result = start;
1392 goto share;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393 }
1394
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 /* obtain the address at which to make a shared mapping
1396 * - this is the hook for quasi-memory character devices to
1397 * tell us the location of a shared mapping
1398 */
David Howells645d83c2009-09-24 15:13:10 +01001399 if (capabilities & BDI_CAP_MAP_DIRECT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400 addr = file->f_op->get_unmapped_area(file, addr, len,
1401 pgoff, flags);
Namhyung Kimbb005a52011-05-24 17:11:27 -07001402 if (IS_ERR_VALUE(addr)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 ret = addr;
Namhyung Kimbb005a52011-05-24 17:11:27 -07001404 if (ret != -ENOSYS)
David Howells8feae132009-01-08 12:04:47 +00001405 goto error_just_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406
1407 /* the driver refused to tell us where to site
1408 * the mapping so we'll have to attempt to copy
1409 * it */
Namhyung Kimbb005a52011-05-24 17:11:27 -07001410 ret = -ENODEV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 if (!(capabilities & BDI_CAP_MAP_COPY))
David Howells8feae132009-01-08 12:04:47 +00001412 goto error_just_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413
1414 capabilities &= ~BDI_CAP_MAP_DIRECT;
David Howells8feae132009-01-08 12:04:47 +00001415 } else {
1416 vma->vm_start = region->vm_start = addr;
1417 vma->vm_end = region->vm_end = addr + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418 }
1419 }
1420 }
1421
David Howells8feae132009-01-08 12:04:47 +00001422 vma->vm_region = region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423
David Howells645d83c2009-09-24 15:13:10 +01001424 /* set up the mapping
1425 * - the region is filled in if BDI_CAP_MAP_DIRECT is still set
1426 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 if (file && vma->vm_flags & VM_SHARED)
David Howells8feae132009-01-08 12:04:47 +00001428 ret = do_mmap_shared_file(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429 else
David Howells645d83c2009-09-24 15:13:10 +01001430 ret = do_mmap_private(vma, region, len, capabilities);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431 if (ret < 0)
David Howells645d83c2009-09-24 15:13:10 +01001432 goto error_just_free;
1433 add_nommu_region(region);
David Howells8feae132009-01-08 12:04:47 +00001434
Jie Zhangea637632009-12-14 18:00:02 -08001435 /* clear anonymous mappings that don't ask for uninitialized data */
1436 if (!vma->vm_file && !(flags & MAP_UNINITIALIZED))
1437 memset((void *)region->vm_start, 0,
1438 region->vm_end - region->vm_start);
1439
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 /* okay... we have a mapping; now we have to register it */
David Howells8feae132009-01-08 12:04:47 +00001441 result = vma->vm_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 current->mm->total_vm += len >> PAGE_SHIFT;
1444
David Howells8feae132009-01-08 12:04:47 +00001445share:
1446 add_vma_to_mm(current->mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447
Mike Frysingercfe79c02010-01-06 17:23:23 +00001448 /* we flush the region from the icache only when the first executable
1449 * mapping of it is made */
1450 if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1451 flush_icache_range(region->vm_start, region->vm_end);
1452 region->vm_icache_flushed = true;
1453 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454
Mike Frysingercfe79c02010-01-06 17:23:23 +00001455 up_write(&nommu_region_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456
David Howells8feae132009-01-08 12:04:47 +00001457 kleave(" = %lx", result);
1458 return result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459
David Howells8feae132009-01-08 12:04:47 +00001460error_just_free:
1461 up_write(&nommu_region_sem);
1462error:
David Howells89a86402009-10-30 13:13:26 +00001463 if (region->vm_file)
1464 fput(region->vm_file);
David Howells8feae132009-01-08 12:04:47 +00001465 kmem_cache_free(vm_region_jar, region);
David Howells89a86402009-10-30 13:13:26 +00001466 if (vma->vm_file)
1467 fput(vma->vm_file);
David Howells8feae132009-01-08 12:04:47 +00001468 kmem_cache_free(vm_area_cachep, vma);
1469 kleave(" = %d", ret);
1470 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471
David Howells8feae132009-01-08 12:04:47 +00001472sharing_violation:
1473 up_write(&nommu_region_sem);
1474 printk(KERN_WARNING "Attempt to share mismatched mappings\n");
1475 ret = -EINVAL;
1476 goto error;
1477
1478error_getting_vma:
1479 kmem_cache_free(vm_region_jar, region);
1480 printk(KERN_WARNING "Allocation of vma for %lu byte allocation"
1481 " from process %d failed\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 len, current->pid);
David Rientjes7bf02ea2011-05-24 17:11:16 -07001483 show_free_areas(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 return -ENOMEM;
1485
David Howells8feae132009-01-08 12:04:47 +00001486error_getting_region:
1487 printk(KERN_WARNING "Allocation of vm region for %lu byte allocation"
1488 " from process %d failed\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 len, current->pid);
David Rientjes7bf02ea2011-05-24 17:11:16 -07001490 show_free_areas(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 return -ENOMEM;
1492}
Linus Torvalds6be5ceb2012-04-20 17:13:58 -07001493
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001494SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1495 unsigned long, prot, unsigned long, flags,
1496 unsigned long, fd, unsigned long, pgoff)
1497{
1498 struct file *file = NULL;
1499 unsigned long retval = -EBADF;
1500
Al Viro120a7952010-10-30 02:54:44 -04001501 audit_mmap_fd(fd, flags);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001502 if (!(flags & MAP_ANONYMOUS)) {
1503 file = fget(fd);
1504 if (!file)
1505 goto out;
1506 }
1507
1508 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1509
Greg Ungererad1ed292012-06-04 14:29:59 +10001510 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001511
1512 if (file)
1513 fput(file);
1514out:
1515 return retval;
1516}
1517
Christoph Hellwiga4679372010-03-10 15:21:15 -08001518#ifdef __ARCH_WANT_SYS_OLD_MMAP
1519struct mmap_arg_struct {
1520 unsigned long addr;
1521 unsigned long len;
1522 unsigned long prot;
1523 unsigned long flags;
1524 unsigned long fd;
1525 unsigned long offset;
1526};
1527
1528SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1529{
1530 struct mmap_arg_struct a;
1531
1532 if (copy_from_user(&a, arg, sizeof(a)))
1533 return -EFAULT;
1534 if (a.offset & ~PAGE_MASK)
1535 return -EINVAL;
1536
1537 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1538 a.offset >> PAGE_SHIFT);
1539}
1540#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1541
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542/*
David Howells8feae132009-01-08 12:04:47 +00001543 * split a vma into two pieces at address 'addr', a new vma is allocated either
1544 * for the first part or the tail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545 */
David Howells8feae132009-01-08 12:04:47 +00001546int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1547 unsigned long addr, int new_below)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548{
David Howells8feae132009-01-08 12:04:47 +00001549 struct vm_area_struct *new;
1550 struct vm_region *region;
1551 unsigned long npages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552
David Howells8feae132009-01-08 12:04:47 +00001553 kenter("");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554
David Howells779c1022010-01-15 17:01:34 -08001555 /* we're only permitted to split anonymous regions (these should have
1556 * only a single usage on the region) */
1557 if (vma->vm_file)
David Howells8feae132009-01-08 12:04:47 +00001558 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559
David Howells8feae132009-01-08 12:04:47 +00001560 if (mm->map_count >= sysctl_max_map_count)
1561 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562
David Howells8feae132009-01-08 12:04:47 +00001563 region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1564 if (!region)
1565 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566
David Howells8feae132009-01-08 12:04:47 +00001567 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1568 if (!new) {
1569 kmem_cache_free(vm_region_jar, region);
1570 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571 }
David Howells8feae132009-01-08 12:04:47 +00001572
1573 /* most fields are the same, copy all, and then fixup */
1574 *new = *vma;
1575 *region = *vma->vm_region;
1576 new->vm_region = region;
1577
1578 npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1579
1580 if (new_below) {
Paul Mundtdd8632a2009-01-08 12:04:47 +00001581 region->vm_top = region->vm_end = new->vm_end = addr;
David Howells8feae132009-01-08 12:04:47 +00001582 } else {
1583 region->vm_start = new->vm_start = addr;
1584 region->vm_pgoff = new->vm_pgoff += npages;
1585 }
1586
1587 if (new->vm_ops && new->vm_ops->open)
1588 new->vm_ops->open(new);
1589
1590 delete_vma_from_mm(vma);
1591 down_write(&nommu_region_sem);
1592 delete_nommu_region(vma->vm_region);
1593 if (new_below) {
1594 vma->vm_region->vm_start = vma->vm_start = addr;
1595 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1596 } else {
1597 vma->vm_region->vm_end = vma->vm_end = addr;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001598 vma->vm_region->vm_top = addr;
David Howells8feae132009-01-08 12:04:47 +00001599 }
1600 add_nommu_region(vma->vm_region);
1601 add_nommu_region(new->vm_region);
1602 up_write(&nommu_region_sem);
1603 add_vma_to_mm(mm, vma);
1604 add_vma_to_mm(mm, new);
1605 return 0;
1606}
1607
1608/*
1609 * shrink a VMA by removing the specified chunk from either the beginning or
1610 * the end
1611 */
1612static int shrink_vma(struct mm_struct *mm,
1613 struct vm_area_struct *vma,
1614 unsigned long from, unsigned long to)
1615{
1616 struct vm_region *region;
1617
1618 kenter("");
1619
1620 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1621 * and list */
1622 delete_vma_from_mm(vma);
1623 if (from > vma->vm_start)
1624 vma->vm_end = from;
1625 else
1626 vma->vm_start = to;
1627 add_vma_to_mm(mm, vma);
1628
1629 /* cut the backing region down to size */
1630 region = vma->vm_region;
David Howells1e2ae592010-01-15 17:01:33 -08001631 BUG_ON(region->vm_usage != 1);
David Howells8feae132009-01-08 12:04:47 +00001632
1633 down_write(&nommu_region_sem);
1634 delete_nommu_region(region);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001635 if (from > region->vm_start) {
1636 to = region->vm_top;
1637 region->vm_top = region->vm_end = from;
1638 } else {
David Howells8feae132009-01-08 12:04:47 +00001639 region->vm_start = to;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001640 }
David Howells8feae132009-01-08 12:04:47 +00001641 add_nommu_region(region);
1642 up_write(&nommu_region_sem);
1643
1644 free_page_series(from, to);
1645 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646}
1647
David Howells30340972006-09-27 01:50:20 -07001648/*
1649 * release a mapping
David Howells8feae132009-01-08 12:04:47 +00001650 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1651 * VMA, though it need not cover the whole VMA
David Howells30340972006-09-27 01:50:20 -07001652 */
David Howells8feae132009-01-08 12:04:47 +00001653int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654{
David Howells8feae132009-01-08 12:04:47 +00001655 struct vm_area_struct *vma;
Bob Liuf67d9b12011-05-24 17:12:56 -07001656 unsigned long end;
David Howells8feae132009-01-08 12:04:47 +00001657 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658
David Howells8feae132009-01-08 12:04:47 +00001659 kenter(",%lx,%zx", start, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660
Bob Liuf67d9b12011-05-24 17:12:56 -07001661 len = PAGE_ALIGN(len);
David Howells8feae132009-01-08 12:04:47 +00001662 if (len == 0)
1663 return -EINVAL;
1664
Bob Liuf67d9b12011-05-24 17:12:56 -07001665 end = start + len;
1666
David Howells8feae132009-01-08 12:04:47 +00001667 /* find the first potentially overlapping VMA */
1668 vma = find_vma(mm, start);
1669 if (!vma) {
Choi Gi-yongac714902014-04-07 15:37:36 -07001670 static int limit;
David Howells33e5d7692009-04-02 16:56:32 -07001671 if (limit < 5) {
1672 printk(KERN_WARNING
1673 "munmap of memory not mmapped by process %d"
1674 " (%s): 0x%lx-0x%lx\n",
1675 current->pid, current->comm,
1676 start, start + len - 1);
1677 limit++;
1678 }
David Howells8feae132009-01-08 12:04:47 +00001679 return -EINVAL;
David Howells30340972006-09-27 01:50:20 -07001680 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681
David Howells8feae132009-01-08 12:04:47 +00001682 /* we're allowed to split an anonymous VMA but not a file-backed one */
1683 if (vma->vm_file) {
1684 do {
1685 if (start > vma->vm_start) {
1686 kleave(" = -EINVAL [miss]");
1687 return -EINVAL;
1688 }
1689 if (end == vma->vm_end)
1690 goto erase_whole_vma;
Namhyung Kimd75a3102011-05-24 17:11:25 -07001691 vma = vma->vm_next;
1692 } while (vma);
David Howells8feae132009-01-08 12:04:47 +00001693 kleave(" = -EINVAL [split file]");
1694 return -EINVAL;
1695 } else {
1696 /* the chunk must be a subset of the VMA found */
1697 if (start == vma->vm_start && end == vma->vm_end)
1698 goto erase_whole_vma;
1699 if (start < vma->vm_start || end > vma->vm_end) {
1700 kleave(" = -EINVAL [superset]");
1701 return -EINVAL;
1702 }
1703 if (start & ~PAGE_MASK) {
1704 kleave(" = -EINVAL [unaligned start]");
1705 return -EINVAL;
1706 }
1707 if (end != vma->vm_end && end & ~PAGE_MASK) {
1708 kleave(" = -EINVAL [unaligned split]");
1709 return -EINVAL;
1710 }
1711 if (start != vma->vm_start && end != vma->vm_end) {
1712 ret = split_vma(mm, vma, start, 1);
1713 if (ret < 0) {
1714 kleave(" = %d [split]", ret);
1715 return ret;
1716 }
1717 }
1718 return shrink_vma(mm, vma, start, end);
1719 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720
David Howells8feae132009-01-08 12:04:47 +00001721erase_whole_vma:
1722 delete_vma_from_mm(vma);
1723 delete_vma(mm, vma);
1724 kleave(" = 0");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 return 0;
1726}
Paul Mundtb5073172007-07-21 04:37:25 -07001727EXPORT_SYMBOL(do_munmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728
Al Virobfce2812012-04-20 21:57:04 -04001729int vm_munmap(unsigned long addr, size_t len)
David Howells30340972006-09-27 01:50:20 -07001730{
Al Virobfce2812012-04-20 21:57:04 -04001731 struct mm_struct *mm = current->mm;
David Howells30340972006-09-27 01:50:20 -07001732 int ret;
David Howells30340972006-09-27 01:50:20 -07001733
1734 down_write(&mm->mmap_sem);
1735 ret = do_munmap(mm, addr, len);
1736 up_write(&mm->mmap_sem);
1737 return ret;
1738}
Linus Torvaldsa46ef992012-04-20 16:20:01 -07001739EXPORT_SYMBOL(vm_munmap);
1740
1741SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1742{
Al Virobfce2812012-04-20 21:57:04 -04001743 return vm_munmap(addr, len);
Linus Torvaldsa46ef992012-04-20 16:20:01 -07001744}
David Howells30340972006-09-27 01:50:20 -07001745
1746/*
David Howells8feae132009-01-08 12:04:47 +00001747 * release all the mappings made in a process's VM space
David Howells30340972006-09-27 01:50:20 -07001748 */
David Howells8feae132009-01-08 12:04:47 +00001749void exit_mmap(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750{
David Howells8feae132009-01-08 12:04:47 +00001751 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752
David Howells8feae132009-01-08 12:04:47 +00001753 if (!mm)
1754 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755
David Howells8feae132009-01-08 12:04:47 +00001756 kenter("");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757
David Howells8feae132009-01-08 12:04:47 +00001758 mm->total_vm = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759
David Howells8feae132009-01-08 12:04:47 +00001760 while ((vma = mm->mmap)) {
1761 mm->mmap = vma->vm_next;
1762 delete_vma_from_mm(vma);
1763 delete_vma(mm, vma);
Steven J. Magnani04c34962010-11-24 12:56:54 -08001764 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765 }
David Howells8feae132009-01-08 12:04:47 +00001766
1767 kleave("");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768}
1769
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07001770unsigned long vm_brk(unsigned long addr, unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771{
1772 return -ENOMEM;
1773}
1774
1775/*
David Howells6fa5f802006-09-27 01:50:21 -07001776 * expand (or shrink) an existing mapping, potentially moving it at the same
1777 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778 *
David Howells6fa5f802006-09-27 01:50:21 -07001779 * under NOMMU conditions, we only permit changing a mapping's size, and only
David Howells8feae132009-01-08 12:04:47 +00001780 * as long as it stays within the region allocated by do_mmap_private() and the
1781 * block is not shareable
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782 *
David Howells6fa5f802006-09-27 01:50:21 -07001783 * MREMAP_FIXED is not supported under NOMMU conditions
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784 */
Al Viro4b377ba2013-03-04 10:47:59 -05001785static unsigned long do_mremap(unsigned long addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786 unsigned long old_len, unsigned long new_len,
1787 unsigned long flags, unsigned long new_addr)
1788{
David Howells6fa5f802006-09-27 01:50:21 -07001789 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790
1791 /* insanity checks first */
Bob Liuf67d9b12011-05-24 17:12:56 -07001792 old_len = PAGE_ALIGN(old_len);
1793 new_len = PAGE_ALIGN(new_len);
David Howells8feae132009-01-08 12:04:47 +00001794 if (old_len == 0 || new_len == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795 return (unsigned long) -EINVAL;
1796
David Howells8feae132009-01-08 12:04:47 +00001797 if (addr & ~PAGE_MASK)
1798 return -EINVAL;
1799
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800 if (flags & MREMAP_FIXED && new_addr != addr)
1801 return (unsigned long) -EINVAL;
1802
David Howells8feae132009-01-08 12:04:47 +00001803 vma = find_vma_exact(current->mm, addr, old_len);
David Howells6fa5f802006-09-27 01:50:21 -07001804 if (!vma)
1805 return (unsigned long) -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806
David Howells6fa5f802006-09-27 01:50:21 -07001807 if (vma->vm_end != vma->vm_start + old_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808 return (unsigned long) -EFAULT;
1809
David Howells6fa5f802006-09-27 01:50:21 -07001810 if (vma->vm_flags & VM_MAYSHARE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811 return (unsigned long) -EPERM;
1812
David Howells8feae132009-01-08 12:04:47 +00001813 if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814 return (unsigned long) -ENOMEM;
1815
1816 /* all checks complete - do it */
David Howells6fa5f802006-09-27 01:50:21 -07001817 vma->vm_end = vma->vm_start + new_len;
David Howells6fa5f802006-09-27 01:50:21 -07001818 return vma->vm_start;
1819}
1820
Heiko Carstens6a6160a2009-01-14 14:14:15 +01001821SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1822 unsigned long, new_len, unsigned long, flags,
1823 unsigned long, new_addr)
David Howells6fa5f802006-09-27 01:50:21 -07001824{
1825 unsigned long ret;
1826
1827 down_write(&current->mm->mmap_sem);
1828 ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1829 up_write(&current->mm->mmap_sem);
1830 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831}
1832
Michel Lespinasse240aade2013-02-22 16:35:56 -08001833struct page *follow_page_mask(struct vm_area_struct *vma,
1834 unsigned long address, unsigned int flags,
1835 unsigned int *page_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836{
Michel Lespinasse240aade2013-02-22 16:35:56 -08001837 *page_mask = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838 return NULL;
1839}
1840
Bob Liu8f3b1322011-07-08 15:39:46 -07001841int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1842 unsigned long pfn, unsigned long size, pgprot_t prot)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843{
Bob Liu8f3b1322011-07-08 15:39:46 -07001844 if (addr != (pfn << PAGE_SHIFT))
1845 return -EINVAL;
1846
Konstantin Khlebnikov314e51b2012-10-08 16:29:02 -07001847 vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
Greg Ungerer66aa2b42005-09-12 11:18:10 +10001848 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849}
Luke Yang22c4af42006-07-14 00:24:09 -07001850EXPORT_SYMBOL(remap_pfn_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851
Linus Torvalds3c0b9de2013-04-27 13:25:38 -07001852int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)
1853{
1854 unsigned long pfn = start >> PAGE_SHIFT;
1855 unsigned long vm_len = vma->vm_end - vma->vm_start;
1856
1857 pfn += vma->vm_pgoff;
1858 return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);
1859}
1860EXPORT_SYMBOL(vm_iomap_memory);
1861
Paul Mundtf905bc42008-02-04 22:29:59 -08001862int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1863 unsigned long pgoff)
1864{
1865 unsigned int size = vma->vm_end - vma->vm_start;
1866
1867 if (!(vma->vm_flags & VM_USERMAP))
1868 return -EINVAL;
1869
1870 vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1871 vma->vm_end = vma->vm_start + size;
1872
1873 return 0;
1874}
1875EXPORT_SYMBOL(remap_vmalloc_range);
1876
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1878 unsigned long len, unsigned long pgoff, unsigned long flags)
1879{
1880 return -ENOMEM;
1881}
1882
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883void unmap_mapping_range(struct address_space *mapping,
1884 loff_t const holebegin, loff_t const holelen,
1885 int even_cows)
1886{
1887}
Luke Yang22c4af42006-07-14 00:24:09 -07001888EXPORT_SYMBOL(unmap_mapping_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889
1890/*
1891 * Check that a process has enough memory to allocate a new virtual
1892 * mapping. 0 means there is enough memory for the allocation to
1893 * succeed and -ENOMEM implies there is not.
1894 *
1895 * We currently support three overcommit policies, which are set via the
1896 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
1897 *
1898 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1899 * Additional code 2002 Jul 20 by Robert Love.
1900 *
1901 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1902 *
1903 * Note this is a helper function intended to be used by LSMs which
1904 * wish to use this logic.
1905 */
Alan Cox34b4e4a2007-08-22 14:01:28 -07001906int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907{
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07001908 unsigned long free, allowed, reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909
1910 vm_acct_memory(pages);
1911
1912 /*
1913 * Sometimes we want to use more memory than we have
1914 */
1915 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1916 return 0;
1917
1918 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
Dmitry Finkc15bef32011-07-25 17:12:19 -07001919 free = global_page_state(NR_FREE_PAGES);
1920 free += global_page_state(NR_FILE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921
Dmitry Finkc15bef32011-07-25 17:12:19 -07001922 /*
1923 * shmem pages shouldn't be counted as free in this
1924 * case, they can't be purged, only swapped out, and
1925 * that won't affect the overall amount of available
1926 * memory in the system.
1927 */
1928 free -= global_page_state(NR_SHMEM);
1929
Shaohua Liec8acf22013-02-22 16:34:38 -08001930 free += get_nr_swap_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931
1932 /*
1933 * Any slabs which are created with the
1934 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1935 * which are reclaimable, under pressure. The dentry
1936 * cache and most inode caches should fall into this
1937 */
Christoph Lameter972d1a72006-09-25 23:31:51 -07001938 free += global_page_state(NR_SLAB_RECLAIMABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939
1940 /*
Dmitry Finkc15bef32011-07-25 17:12:19 -07001941 * Leave reserved pages. The pages are not for anonymous pages.
1942 */
1943 if (free <= totalreserve_pages)
1944 goto error;
1945 else
1946 free -= totalreserve_pages;
1947
1948 /*
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07001949 * Reserve some for root
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950 */
1951 if (!cap_sys_admin)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07001952 free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953
1954 if (free > pages)
1955 return 0;
1956
Hideo AOKId5ddc792006-04-10 22:53:01 -07001957 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958 }
1959
Jerome Marchand00619bc2013-11-12 15:08:31 -08001960 allowed = vm_commit_limit();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961 /*
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07001962 * Reserve some 3% for root
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963 */
1964 if (!cap_sys_admin)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07001965 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07001967 /*
1968 * Don't let a single process grow so big a user can't recover
1969 */
1970 if (mm) {
1971 reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
1972 allowed -= min(mm->total_vm / 32, reserve);
1973 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07001975 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976 return 0;
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07001977
Hideo AOKId5ddc792006-04-10 22:53:01 -07001978error:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979 vm_unacct_memory(pages);
1980
1981 return -ENOMEM;
1982}
1983
Nick Piggind0217ac2007-07-19 01:47:03 -07001984int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
David Howellsb0e15192006-01-06 00:11:42 -08001985{
1986 BUG();
Nick Piggind0217ac2007-07-19 01:47:03 -07001987 return 0;
David Howellsb0e15192006-01-06 00:11:42 -08001988}
Paul Mundtb5073172007-07-21 04:37:25 -07001989EXPORT_SYMBOL(filemap_fault);
David Howells0ec76a12006-09-27 01:50:15 -07001990
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07001991void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf)
1992{
1993 BUG();
1994}
1995EXPORT_SYMBOL(filemap_map_pages);
1996
Konstantin Khlebnikov0b173bc2012-10-08 16:28:46 -07001997int generic_file_remap_pages(struct vm_area_struct *vma, unsigned long addr,
1998 unsigned long size, pgoff_t pgoff)
1999{
2000 BUG();
2001 return 0;
2002}
2003EXPORT_SYMBOL(generic_file_remap_pages);
2004
Mike Frysingerf55f1992011-03-29 14:05:12 +01002005static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
2006 unsigned long addr, void *buf, int len, int write)
David Howells0ec76a12006-09-27 01:50:15 -07002007{
David Howells0ec76a12006-09-27 01:50:15 -07002008 struct vm_area_struct *vma;
David Howells0ec76a12006-09-27 01:50:15 -07002009
2010 down_read(&mm->mmap_sem);
2011
2012 /* the access must start within one of the target process's mappings */
David Howells0159b142006-09-27 01:50:16 -07002013 vma = find_vma(mm, addr);
2014 if (vma) {
David Howells0ec76a12006-09-27 01:50:15 -07002015 /* don't overrun this mapping */
2016 if (addr + len >= vma->vm_end)
2017 len = vma->vm_end - addr;
2018
2019 /* only read or write mappings where it is permitted */
David Howellsd00c7b992006-09-27 01:50:19 -07002020 if (write && vma->vm_flags & VM_MAYWRITE)
Jie Zhang79597222010-01-06 17:23:28 +00002021 copy_to_user_page(vma, NULL, addr,
2022 (void *) addr, buf, len);
David Howellsd00c7b992006-09-27 01:50:19 -07002023 else if (!write && vma->vm_flags & VM_MAYREAD)
Jie Zhang79597222010-01-06 17:23:28 +00002024 copy_from_user_page(vma, NULL, addr,
2025 buf, (void *) addr, len);
David Howells0ec76a12006-09-27 01:50:15 -07002026 else
2027 len = 0;
2028 } else {
2029 len = 0;
2030 }
2031
2032 up_read(&mm->mmap_sem);
Mike Frysingerf55f1992011-03-29 14:05:12 +01002033
2034 return len;
2035}
2036
2037/**
2038 * @access_remote_vm - access another process' address space
2039 * @mm: the mm_struct of the target address space
2040 * @addr: start address to access
2041 * @buf: source or destination buffer
2042 * @len: number of bytes to transfer
2043 * @write: whether the access is a write
2044 *
2045 * The caller must hold a reference on @mm.
2046 */
2047int access_remote_vm(struct mm_struct *mm, unsigned long addr,
2048 void *buf, int len, int write)
2049{
2050 return __access_remote_vm(NULL, mm, addr, buf, len, write);
2051}
2052
2053/*
2054 * Access another process' address space.
2055 * - source/target buffer must be kernel space
2056 */
2057int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
2058{
2059 struct mm_struct *mm;
2060
2061 if (addr + len < addr)
2062 return 0;
2063
2064 mm = get_task_mm(tsk);
2065 if (!mm)
2066 return 0;
2067
2068 len = __access_remote_vm(tsk, mm, addr, buf, len, write);
2069
David Howells0ec76a12006-09-27 01:50:15 -07002070 mmput(mm);
2071 return len;
2072}
David Howells7e660872010-01-15 17:01:39 -08002073
2074/**
2075 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
2076 * @inode: The inode to check
2077 * @size: The current filesize of the inode
2078 * @newsize: The proposed filesize of the inode
2079 *
2080 * Check the shared mappings on an inode on behalf of a shrinking truncate to
2081 * make sure that that any outstanding VMAs aren't broken and then shrink the
2082 * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
2083 * automatically grant mappings that are too large.
2084 */
2085int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
2086 size_t newsize)
2087{
2088 struct vm_area_struct *vma;
David Howells7e660872010-01-15 17:01:39 -08002089 struct vm_region *region;
2090 pgoff_t low, high;
2091 size_t r_size, r_top;
2092
2093 low = newsize >> PAGE_SHIFT;
2094 high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
2095
2096 down_write(&nommu_region_sem);
David Howells918e5562012-02-23 13:50:35 +00002097 mutex_lock(&inode->i_mapping->i_mmap_mutex);
David Howells7e660872010-01-15 17:01:39 -08002098
2099 /* search for VMAs that fall within the dead zone */
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -07002100 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) {
David Howells7e660872010-01-15 17:01:39 -08002101 /* found one - only interested if it's shared out of the page
2102 * cache */
2103 if (vma->vm_flags & VM_SHARED) {
David Howells918e5562012-02-23 13:50:35 +00002104 mutex_unlock(&inode->i_mapping->i_mmap_mutex);
David Howells7e660872010-01-15 17:01:39 -08002105 up_write(&nommu_region_sem);
2106 return -ETXTBSY; /* not quite true, but near enough */
2107 }
2108 }
2109
2110 /* reduce any regions that overlap the dead zone - if in existence,
2111 * these will be pointed to by VMAs that don't overlap the dead zone
2112 *
2113 * we don't check for any regions that start beyond the EOF as there
2114 * shouldn't be any
2115 */
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -07002116 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap,
2117 0, ULONG_MAX) {
David Howells7e660872010-01-15 17:01:39 -08002118 if (!(vma->vm_flags & VM_SHARED))
2119 continue;
2120
2121 region = vma->vm_region;
2122 r_size = region->vm_top - region->vm_start;
2123 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
2124
2125 if (r_top > newsize) {
2126 region->vm_top -= r_top - newsize;
2127 if (region->vm_end > region->vm_top)
2128 region->vm_end = region->vm_top;
2129 }
2130 }
2131
David Howells918e5562012-02-23 13:50:35 +00002132 mutex_unlock(&inode->i_mapping->i_mmap_mutex);
David Howells7e660872010-01-15 17:01:39 -08002133 up_write(&nommu_region_sem);
2134 return 0;
2135}
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -07002136
2137/*
2138 * Initialise sysctl_user_reserve_kbytes.
2139 *
2140 * This is intended to prevent a user from starting a single memory hogging
2141 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
2142 * mode.
2143 *
2144 * The default value is min(3% of free memory, 128MB)
2145 * 128MB is enough to recover with sshd/login, bash, and top/kill.
2146 */
2147static int __meminit init_user_reserve(void)
2148{
2149 unsigned long free_kbytes;
2150
2151 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
2152
2153 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
2154 return 0;
2155}
2156module_init(init_user_reserve)
Andrew Shewmaker4eeab4f2013-04-29 15:08:11 -07002157
2158/*
2159 * Initialise sysctl_admin_reserve_kbytes.
2160 *
2161 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
2162 * to log in and kill a memory hogging process.
2163 *
2164 * Systems with more than 256MB will reserve 8MB, enough to recover
2165 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
2166 * only reserve 3% of free pages by default.
2167 */
2168static int __meminit init_admin_reserve(void)
2169{
2170 unsigned long free_kbytes;
2171
2172 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
2173
2174 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
2175 return 0;
2176}
2177module_init(init_admin_reserve)