blob: 634193324a6b128c9680f4d45bfa4ed6e48fea20 [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
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040016#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/mm.h>
18#include <linux/mman.h>
19#include <linux/swap.h>
20#include <linux/file.h>
21#include <linux/highmem.h>
22#include <linux/pagemap.h>
23#include <linux/slab.h>
24#include <linux/vmalloc.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/blkdev.h>
26#include <linux/backing-dev.h>
27#include <linux/mount.h>
28#include <linux/personality.h>
29#include <linux/security.h>
30#include <linux/syscalls.h>
Al Viro120a7952010-10-30 02:54:44 -040031#include <linux/audit.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070032
33#include <asm/uaccess.h>
34#include <asm/tlb.h>
35#include <asm/tlbflush.h>
Bernd Schmidteb8cdec2009-09-21 17:03:57 -070036#include <asm/mmu_context.h>
David Howells8feae132009-01-08 12:04:47 +000037#include "internal.h"
38
David Howells8feae132009-01-08 12:04:47 +000039#if 0
40#define kenter(FMT, ...) \
41 printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
42#define kleave(FMT, ...) \
43 printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
44#define kdebug(FMT, ...) \
45 printk(KERN_DEBUG "xxx" FMT"yyy\n", ##__VA_ARGS__)
46#else
47#define kenter(FMT, ...) \
48 no_printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
49#define kleave(FMT, ...) \
50 no_printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
51#define kdebug(FMT, ...) \
52 no_printk(KERN_DEBUG FMT"\n", ##__VA_ARGS__)
53#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -070054
55void *high_memory;
56struct page *mem_map;
57unsigned long max_mapnr;
58unsigned long num_physpages;
Hugh Dickins4266c972009-09-23 17:05:53 +010059unsigned long highest_memmap_pfn;
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -070060struct percpu_counter vm_committed_as;
Linus Torvalds1da177e2005-04-16 15:20:36 -070061int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
62int sysctl_overcommit_ratio = 50; /* default is 50% */
63int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
David Howellsfc4d5c22009-05-06 16:03:05 -070064int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -070065int heap_stack_gap = 0;
66
David Howells33e5d7692009-04-02 16:56:32 -070067atomic_long_t mmap_pages_allocated;
David Howells8feae132009-01-08 12:04:47 +000068
Linus Torvalds1da177e2005-04-16 15:20:36 -070069EXPORT_SYMBOL(mem_map);
Wu, Bryan6a04de62007-04-11 23:28:47 -070070EXPORT_SYMBOL(num_physpages);
Linus Torvalds1da177e2005-04-16 15:20:36 -070071
David Howells8feae132009-01-08 12:04:47 +000072/* list of mapped, potentially shareable regions */
73static struct kmem_cache *vm_region_jar;
74struct rb_root nommu_region_tree = RB_ROOT;
75DECLARE_RWSEM(nommu_region_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -070076
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +040077const struct vm_operations_struct generic_file_vm_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -070078};
79
80/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070081 * Return the total memory allocated for this pointer, not
82 * just what the caller asked for.
83 *
84 * Doesn't have to be accurate, i.e. may have races.
85 */
86unsigned int kobjsize(const void *objp)
87{
88 struct page *page;
89
Michael Hennerich4016a132008-04-28 02:13:38 -070090 /*
91 * If the object we have should not have ksize performed on it,
92 * return size of 0
93 */
Paul Mundt5a1603b2008-06-12 16:29:55 +090094 if (!objp || !virt_addr_valid(objp))
Paul Mundt6cfd53fc2008-06-05 22:46:08 -070095 return 0;
96
97 page = virt_to_head_page(objp);
Paul Mundt6cfd53fc2008-06-05 22:46:08 -070098
99 /*
100 * If the allocator sets PageSlab, we know the pointer came from
101 * kmalloc().
102 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103 if (PageSlab(page))
104 return ksize(objp);
105
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700106 /*
Paul Mundtab2e83e2009-01-08 12:04:48 +0000107 * If it's not a compound page, see if we have a matching VMA
108 * region. This test is intentionally done in reverse order,
109 * so if there's no VMA, we still fall through and hand back
110 * PAGE_SIZE for 0-order pages.
111 */
112 if (!PageCompound(page)) {
113 struct vm_area_struct *vma;
114
115 vma = find_vma(current->mm, (unsigned long)objp);
116 if (vma)
117 return vma->vm_end - vma->vm_start;
118 }
119
120 /*
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700121 * The ksize() function is only guaranteed to work for pointers
Paul Mundt5a1603b2008-06-12 16:29:55 +0900122 * returned by kmalloc(). So handle arbitrary pointers here.
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700123 */
Paul Mundt5a1603b2008-06-12 16:29:55 +0900124 return PAGE_SIZE << compound_order(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125}
126
Nick Pigginb291f002008-10-18 20:26:44 -0700127int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
Hugh Dickins4266c972009-09-23 17:05:53 +0100128 unsigned long start, int nr_pages, unsigned int foll_flags,
Michel Lespinasse53a77062011-01-13 15:46:14 -0800129 struct page **pages, struct vm_area_struct **vmas,
130 int *retry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131{
Sonic Zhang910e46d2006-09-27 01:50:17 -0700132 struct vm_area_struct *vma;
David Howells7b4d5b82006-09-27 01:50:18 -0700133 unsigned long vm_flags;
134 int i;
135
136 /* calculate required read or write permissions.
Hugh Dickins58fa8792009-09-21 17:03:31 -0700137 * If FOLL_FORCE is set, we only require the "MAY" flags.
David Howells7b4d5b82006-09-27 01:50:18 -0700138 */
Hugh Dickins58fa8792009-09-21 17:03:31 -0700139 vm_flags = (foll_flags & FOLL_WRITE) ?
140 (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
141 vm_flags &= (foll_flags & FOLL_FORCE) ?
142 (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143
Peter Zijlstra9d737772009-06-25 11:58:55 +0200144 for (i = 0; i < nr_pages; i++) {
David Howells7561e8c2010-03-25 16:48:38 +0000145 vma = find_vma(mm, start);
David Howells7b4d5b82006-09-27 01:50:18 -0700146 if (!vma)
147 goto finish_or_fault;
148
149 /* protect what we can, including chardevs */
Hugh Dickins1c3aff12009-09-21 17:03:24 -0700150 if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
151 !(vm_flags & vma->vm_flags))
David Howells7b4d5b82006-09-27 01:50:18 -0700152 goto finish_or_fault;
Sonic Zhang910e46d2006-09-27 01:50:17 -0700153
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 if (pages) {
155 pages[i] = virt_to_page(start);
156 if (pages[i])
157 page_cache_get(pages[i]);
158 }
159 if (vmas)
Sonic Zhang910e46d2006-09-27 01:50:17 -0700160 vmas[i] = vma;
David Howellse1ee65d2010-03-25 16:48:44 +0000161 start = (start + PAGE_SIZE) & PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 }
David Howells7b4d5b82006-09-27 01:50:18 -0700163
164 return i;
165
166finish_or_fault:
167 return i ? : -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168}
Nick Pigginb291f002008-10-18 20:26:44 -0700169
Nick Pigginb291f002008-10-18 20:26:44 -0700170/*
171 * get a list of pages in an address range belonging to the specified process
172 * and indicate the VMA that covers each page
173 * - this is potentially dodgy as we may end incrementing the page count of a
174 * slab page or a secondary page from a compound page
175 * - don't permit access to VMAs that don't support it, such as I/O mappings
176 */
177int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
Peter Zijlstra9d737772009-06-25 11:58:55 +0200178 unsigned long start, int nr_pages, int write, int force,
Nick Pigginb291f002008-10-18 20:26:44 -0700179 struct page **pages, struct vm_area_struct **vmas)
180{
181 int flags = 0;
182
183 if (write)
Hugh Dickins58fa8792009-09-21 17:03:31 -0700184 flags |= FOLL_WRITE;
Nick Pigginb291f002008-10-18 20:26:44 -0700185 if (force)
Hugh Dickins58fa8792009-09-21 17:03:31 -0700186 flags |= FOLL_FORCE;
Nick Pigginb291f002008-10-18 20:26:44 -0700187
Michel Lespinasse53a77062011-01-13 15:46:14 -0800188 return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas,
189 NULL);
Nick Pigginb291f002008-10-18 20:26:44 -0700190}
Greg Ungerer66aa2b42005-09-12 11:18:10 +1000191EXPORT_SYMBOL(get_user_pages);
192
Paul Mundtdfc2f912009-06-26 04:31:57 +0900193/**
194 * follow_pfn - look up PFN at a user virtual address
195 * @vma: memory mapping
196 * @address: user virtual address
197 * @pfn: location to store found PFN
198 *
199 * Only IO mappings and raw PFN mappings are allowed.
200 *
201 * Returns zero and the pfn at @pfn on success, -ve otherwise.
202 */
203int follow_pfn(struct vm_area_struct *vma, unsigned long address,
204 unsigned long *pfn)
205{
206 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
207 return -EINVAL;
208
209 *pfn = address >> PAGE_SHIFT;
210 return 0;
211}
212EXPORT_SYMBOL(follow_pfn);
213
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214DEFINE_RWLOCK(vmlist_lock);
215struct vm_struct *vmlist;
216
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800217void vfree(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218{
219 kfree(addr);
220}
Paul Mundtb5073172007-07-21 04:37:25 -0700221EXPORT_SYMBOL(vfree);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222
Al Virodd0fc662005-10-07 07:46:04 +0100223void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224{
225 /*
Robert P. J. Day85186092007-10-19 23:11:38 +0200226 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
227 * returns only a logical address.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 */
Nick Piggin84097512006-03-22 00:08:34 -0800229 return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230}
Paul Mundtb5073172007-07-21 04:37:25 -0700231EXPORT_SYMBOL(__vmalloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232
Paul Mundtf905bc42008-02-04 22:29:59 -0800233void *vmalloc_user(unsigned long size)
234{
235 void *ret;
236
237 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
238 PAGE_KERNEL);
239 if (ret) {
240 struct vm_area_struct *vma;
241
242 down_write(&current->mm->mmap_sem);
243 vma = find_vma(current->mm, (unsigned long)ret);
244 if (vma)
245 vma->vm_flags |= VM_USERMAP;
246 up_write(&current->mm->mmap_sem);
247 }
248
249 return ret;
250}
251EXPORT_SYMBOL(vmalloc_user);
252
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800253struct page *vmalloc_to_page(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254{
255 return virt_to_page(addr);
256}
Paul Mundtb5073172007-07-21 04:37:25 -0700257EXPORT_SYMBOL(vmalloc_to_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800259unsigned long vmalloc_to_pfn(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260{
261 return page_to_pfn(virt_to_page(addr));
262}
Paul Mundtb5073172007-07-21 04:37:25 -0700263EXPORT_SYMBOL(vmalloc_to_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264
265long vread(char *buf, char *addr, unsigned long count)
266{
267 memcpy(buf, addr, count);
268 return count;
269}
270
271long vwrite(char *buf, char *addr, unsigned long count)
272{
273 /* Don't allow overflow */
274 if ((unsigned long) addr + count < count)
275 count = -(unsigned long) addr;
276
277 memcpy(addr, buf, count);
278 return(count);
279}
280
281/*
282 * vmalloc - allocate virtually continguos memory
283 *
284 * @size: allocation size
285 *
286 * Allocate enough pages to cover @size from the page level
287 * allocator and map them into continguos kernel virtual space.
288 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +0200289 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290 * use __vmalloc() instead.
291 */
292void *vmalloc(unsigned long size)
293{
294 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
295}
Andrew Mortonf6138882006-02-28 16:59:18 -0800296EXPORT_SYMBOL(vmalloc);
297
Dave Younge1ca7782010-10-26 14:22:06 -0700298/*
299 * vzalloc - allocate virtually continguos memory with zero fill
300 *
301 * @size: allocation size
302 *
303 * Allocate enough pages to cover @size from the page level
304 * allocator and map them into continguos kernel virtual space.
305 * The memory allocated is set to zero.
306 *
307 * For tight control over page level allocator and protection flags
308 * use __vmalloc() instead.
309 */
310void *vzalloc(unsigned long size)
311{
312 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
313 PAGE_KERNEL);
314}
315EXPORT_SYMBOL(vzalloc);
316
317/**
318 * vmalloc_node - allocate memory on a specific node
319 * @size: allocation size
320 * @node: numa node
321 *
322 * Allocate enough pages to cover @size from the page level
323 * allocator and map them into contiguous kernel virtual space.
324 *
325 * For tight control over page level allocator and protection flags
326 * use __vmalloc() instead.
327 */
Andrew Mortonf6138882006-02-28 16:59:18 -0800328void *vmalloc_node(unsigned long size, int node)
329{
330 return vmalloc(size);
331}
Paul Mundt9a14f652010-12-24 11:50:34 +0900332EXPORT_SYMBOL(vmalloc_node);
Dave Younge1ca7782010-10-26 14:22:06 -0700333
334/**
335 * vzalloc_node - allocate memory on a specific node with zero fill
336 * @size: allocation size
337 * @node: numa node
338 *
339 * Allocate enough pages to cover @size from the page level
340 * allocator and map them into contiguous kernel virtual space.
341 * The memory allocated is set to zero.
342 *
343 * For tight control over page level allocator and protection flags
344 * use __vmalloc() instead.
345 */
346void *vzalloc_node(unsigned long size, int node)
347{
348 return vzalloc(size);
349}
350EXPORT_SYMBOL(vzalloc_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351
Paul Mundt1af446e2008-08-04 16:01:47 +0900352#ifndef PAGE_KERNEL_EXEC
353# define PAGE_KERNEL_EXEC PAGE_KERNEL
354#endif
355
356/**
357 * vmalloc_exec - allocate virtually contiguous, executable memory
358 * @size: allocation size
359 *
360 * Kernel-internal function to allocate enough pages to cover @size
361 * the page level allocator and map them into contiguous and
362 * executable kernel virtual space.
363 *
364 * For tight control over page level allocator and protection flags
365 * use __vmalloc() instead.
366 */
367
368void *vmalloc_exec(unsigned long size)
369{
370 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
371}
372
Paul Mundtb5073172007-07-21 04:37:25 -0700373/**
374 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 * @size: allocation size
376 *
377 * Allocate enough 32bit PA addressable pages to cover @size from the
378 * page level allocator and map them into continguos kernel virtual space.
379 */
380void *vmalloc_32(unsigned long size)
381{
382 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
383}
Paul Mundtb5073172007-07-21 04:37:25 -0700384EXPORT_SYMBOL(vmalloc_32);
385
386/**
387 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
388 * @size: allocation size
389 *
390 * The resulting memory area is 32bit addressable and zeroed so it can be
391 * mapped to userspace without leaking data.
Paul Mundtf905bc42008-02-04 22:29:59 -0800392 *
393 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
394 * remap_vmalloc_range() are permissible.
Paul Mundtb5073172007-07-21 04:37:25 -0700395 */
396void *vmalloc_32_user(unsigned long size)
397{
Paul Mundtf905bc42008-02-04 22:29:59 -0800398 /*
399 * We'll have to sort out the ZONE_DMA bits for 64-bit,
400 * but for now this can simply use vmalloc_user() directly.
401 */
402 return vmalloc_user(size);
Paul Mundtb5073172007-07-21 04:37:25 -0700403}
404EXPORT_SYMBOL(vmalloc_32_user);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405
406void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
407{
408 BUG();
409 return NULL;
410}
Paul Mundtb5073172007-07-21 04:37:25 -0700411EXPORT_SYMBOL(vmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800413void vunmap(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414{
415 BUG();
416}
Paul Mundtb5073172007-07-21 04:37:25 -0700417EXPORT_SYMBOL(vunmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418
Paul Mundteb6434d2009-01-21 17:45:47 +0900419void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
420{
421 BUG();
422 return NULL;
423}
424EXPORT_SYMBOL(vm_map_ram);
425
426void vm_unmap_ram(const void *mem, unsigned int count)
427{
428 BUG();
429}
430EXPORT_SYMBOL(vm_unmap_ram);
431
432void vm_unmap_aliases(void)
433{
434}
435EXPORT_SYMBOL_GPL(vm_unmap_aliases);
436
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437/*
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -0700438 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
439 * have one.
440 */
441void __attribute__((weak)) vmalloc_sync_all(void)
442{
443}
444
Paul Mundt29c185e2010-12-24 12:08:30 +0900445/**
446 * alloc_vm_area - allocate a range of kernel address space
447 * @size: size of the area
448 *
449 * Returns: NULL on failure, vm_struct on success
450 *
451 * This function reserves a range of kernel address space, and
452 * allocates pagetables to map that range. No actual mappings
453 * are created. If the kernel address space is not shared
454 * between processes, it syncs the pagetable across all
455 * processes.
456 */
David Vrabelcd129092011-09-29 16:53:32 +0100457struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
Paul Mundt29c185e2010-12-24 12:08:30 +0900458{
459 BUG();
460 return NULL;
461}
462EXPORT_SYMBOL_GPL(alloc_vm_area);
463
464void free_vm_area(struct vm_struct *area)
465{
466 BUG();
467}
468EXPORT_SYMBOL_GPL(free_vm_area);
469
Paul Mundtb5073172007-07-21 04:37:25 -0700470int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
471 struct page *page)
472{
473 return -EINVAL;
474}
475EXPORT_SYMBOL(vm_insert_page);
476
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -0700477/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 * sys_brk() for the most part doesn't need the global kernel
479 * lock, except when an application is doing something nasty
480 * like trying to un-brk an area that has already been mapped
481 * to a regular file. in this case, the unmapping will need
482 * to invoke file system routines that need the global lock.
483 */
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100484SYSCALL_DEFINE1(brk, unsigned long, brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485{
486 struct mm_struct *mm = current->mm;
487
488 if (brk < mm->start_brk || brk > mm->context.end_brk)
489 return mm->brk;
490
491 if (mm->brk == brk)
492 return mm->brk;
493
494 /*
495 * Always allow shrinking brk
496 */
497 if (brk <= mm->brk) {
498 mm->brk = brk;
499 return brk;
500 }
501
502 /*
503 * Ok, looks good - let it rip.
504 */
Mike Frysingercfe79c02010-01-06 17:23:23 +0000505 flush_icache_range(mm->brk, brk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 return mm->brk = brk;
507}
508
David Howells8feae132009-01-08 12:04:47 +0000509/*
510 * initialise the VMA and region record slabs
511 */
512void __init mmap_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513{
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -0700514 int ret;
515
516 ret = percpu_counter_init(&vm_committed_as, 0);
517 VM_BUG_ON(ret);
David Howells33e5d7692009-04-02 16:56:32 -0700518 vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC);
David Howells8feae132009-01-08 12:04:47 +0000519}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520
David Howells8feae132009-01-08 12:04:47 +0000521/*
522 * validate the region tree
523 * - the caller must hold the region lock
524 */
525#ifdef CONFIG_DEBUG_NOMMU_REGIONS
526static noinline void validate_nommu_regions(void)
527{
528 struct vm_region *region, *last;
529 struct rb_node *p, *lastp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530
David Howells8feae132009-01-08 12:04:47 +0000531 lastp = rb_first(&nommu_region_tree);
532 if (!lastp)
533 return;
534
535 last = rb_entry(lastp, struct vm_region, vm_rb);
David Howells33e5d7692009-04-02 16:56:32 -0700536 BUG_ON(unlikely(last->vm_end <= last->vm_start));
537 BUG_ON(unlikely(last->vm_top < last->vm_end));
David Howells8feae132009-01-08 12:04:47 +0000538
539 while ((p = rb_next(lastp))) {
540 region = rb_entry(p, struct vm_region, vm_rb);
541 last = rb_entry(lastp, struct vm_region, vm_rb);
542
David Howells33e5d7692009-04-02 16:56:32 -0700543 BUG_ON(unlikely(region->vm_end <= region->vm_start));
544 BUG_ON(unlikely(region->vm_top < region->vm_end));
545 BUG_ON(unlikely(region->vm_start < last->vm_top));
David Howells8feae132009-01-08 12:04:47 +0000546
547 lastp = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 }
549}
David Howells8feae132009-01-08 12:04:47 +0000550#else
David Howells33e5d7692009-04-02 16:56:32 -0700551static void validate_nommu_regions(void)
552{
553}
David Howells8feae132009-01-08 12:04:47 +0000554#endif
555
556/*
557 * add a region into the global tree
558 */
559static void add_nommu_region(struct vm_region *region)
560{
561 struct vm_region *pregion;
562 struct rb_node **p, *parent;
563
564 validate_nommu_regions();
565
David Howells8feae132009-01-08 12:04:47 +0000566 parent = NULL;
567 p = &nommu_region_tree.rb_node;
568 while (*p) {
569 parent = *p;
570 pregion = rb_entry(parent, struct vm_region, vm_rb);
571 if (region->vm_start < pregion->vm_start)
572 p = &(*p)->rb_left;
573 else if (region->vm_start > pregion->vm_start)
574 p = &(*p)->rb_right;
575 else if (pregion == region)
576 return;
577 else
578 BUG();
579 }
580
581 rb_link_node(&region->vm_rb, parent, p);
582 rb_insert_color(&region->vm_rb, &nommu_region_tree);
583
584 validate_nommu_regions();
585}
586
587/*
588 * delete a region from the global tree
589 */
590static void delete_nommu_region(struct vm_region *region)
591{
592 BUG_ON(!nommu_region_tree.rb_node);
593
594 validate_nommu_regions();
595 rb_erase(&region->vm_rb, &nommu_region_tree);
596 validate_nommu_regions();
597}
598
599/*
600 * free a contiguous series of pages
601 */
602static void free_page_series(unsigned long from, unsigned long to)
603{
604 for (; from < to; from += PAGE_SIZE) {
605 struct page *page = virt_to_page(from);
606
607 kdebug("- free %lx", from);
David Howells33e5d7692009-04-02 16:56:32 -0700608 atomic_long_dec(&mmap_pages_allocated);
David Howells8feae132009-01-08 12:04:47 +0000609 if (page_count(page) != 1)
David Howells33e5d7692009-04-02 16:56:32 -0700610 kdebug("free page %p: refcount not one: %d",
611 page, page_count(page));
David Howells8feae132009-01-08 12:04:47 +0000612 put_page(page);
613 }
614}
615
616/*
617 * release a reference to a region
David Howells33e5d7692009-04-02 16:56:32 -0700618 * - the caller must hold the region semaphore for writing, which this releases
Paul Mundtdd8632a2009-01-08 12:04:47 +0000619 * - the region may not have been added to the tree yet, in which case vm_top
David Howells8feae132009-01-08 12:04:47 +0000620 * will equal vm_start
621 */
622static void __put_nommu_region(struct vm_region *region)
623 __releases(nommu_region_sem)
624{
David Howells1e2ae592010-01-15 17:01:33 -0800625 kenter("%p{%d}", region, region->vm_usage);
David Howells8feae132009-01-08 12:04:47 +0000626
627 BUG_ON(!nommu_region_tree.rb_node);
628
David Howells1e2ae592010-01-15 17:01:33 -0800629 if (--region->vm_usage == 0) {
Paul Mundtdd8632a2009-01-08 12:04:47 +0000630 if (region->vm_top > region->vm_start)
David Howells8feae132009-01-08 12:04:47 +0000631 delete_nommu_region(region);
632 up_write(&nommu_region_sem);
633
634 if (region->vm_file)
635 fput(region->vm_file);
636
637 /* IO memory and memory shared directly out of the pagecache
638 * from ramfs/tmpfs mustn't be released here */
639 if (region->vm_flags & VM_MAPPED_COPY) {
640 kdebug("free series");
Paul Mundtdd8632a2009-01-08 12:04:47 +0000641 free_page_series(region->vm_start, region->vm_top);
David Howells8feae132009-01-08 12:04:47 +0000642 }
643 kmem_cache_free(vm_region_jar, region);
644 } else {
645 up_write(&nommu_region_sem);
646 }
647}
648
649/*
650 * release a reference to a region
651 */
652static void put_nommu_region(struct vm_region *region)
653{
654 down_write(&nommu_region_sem);
655 __put_nommu_region(region);
656}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657
David Howells30340972006-09-27 01:50:20 -0700658/*
Bernd Schmidteb8cdec2009-09-21 17:03:57 -0700659 * update protection on a vma
660 */
661static void protect_vma(struct vm_area_struct *vma, unsigned long flags)
662{
663#ifdef CONFIG_MPU
664 struct mm_struct *mm = vma->vm_mm;
665 long start = vma->vm_start & PAGE_MASK;
666 while (start < vma->vm_end) {
667 protect_page(mm, start, flags);
668 start += PAGE_SIZE;
669 }
670 update_protections(mm);
671#endif
672}
673
674/*
David Howells30340972006-09-27 01:50:20 -0700675 * add a VMA into a process's mm_struct in the appropriate place in the list
David Howells8feae132009-01-08 12:04:47 +0000676 * and tree and add to the address space's page tree also if not an anonymous
677 * page
David Howells30340972006-09-27 01:50:20 -0700678 * - should be called with mm->mmap_sem held writelocked
679 */
David Howells8feae132009-01-08 12:04:47 +0000680static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
David Howells30340972006-09-27 01:50:20 -0700681{
Namhyung Kim6038def2011-05-24 17:11:22 -0700682 struct vm_area_struct *pvma, *prev;
David Howells8feae132009-01-08 12:04:47 +0000683 struct address_space *mapping;
Namhyung Kim6038def2011-05-24 17:11:22 -0700684 struct rb_node **p, *parent, *rb_prev;
David Howells30340972006-09-27 01:50:20 -0700685
David Howells8feae132009-01-08 12:04:47 +0000686 kenter(",%p", vma);
687
688 BUG_ON(!vma->vm_region);
689
690 mm->map_count++;
691 vma->vm_mm = mm;
692
Bernd Schmidteb8cdec2009-09-21 17:03:57 -0700693 protect_vma(vma, vma->vm_flags);
694
David Howells8feae132009-01-08 12:04:47 +0000695 /* add the VMA to the mapping */
696 if (vma->vm_file) {
697 mapping = vma->vm_file->f_mapping;
698
David Howells918e5562012-02-23 13:50:35 +0000699 mutex_lock(&mapping->i_mmap_mutex);
David Howells8feae132009-01-08 12:04:47 +0000700 flush_dcache_mmap_lock(mapping);
701 vma_prio_tree_insert(vma, &mapping->i_mmap);
702 flush_dcache_mmap_unlock(mapping);
David Howells918e5562012-02-23 13:50:35 +0000703 mutex_unlock(&mapping->i_mmap_mutex);
David Howells8feae132009-01-08 12:04:47 +0000704 }
705
706 /* add the VMA to the tree */
Namhyung Kim6038def2011-05-24 17:11:22 -0700707 parent = rb_prev = NULL;
David Howells8feae132009-01-08 12:04:47 +0000708 p = &mm->mm_rb.rb_node;
709 while (*p) {
710 parent = *p;
711 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
712
713 /* sort by: start addr, end addr, VMA struct addr in that order
714 * (the latter is necessary as we may get identical VMAs) */
715 if (vma->vm_start < pvma->vm_start)
716 p = &(*p)->rb_left;
Namhyung Kim6038def2011-05-24 17:11:22 -0700717 else if (vma->vm_start > pvma->vm_start) {
718 rb_prev = parent;
David Howells8feae132009-01-08 12:04:47 +0000719 p = &(*p)->rb_right;
Namhyung Kim6038def2011-05-24 17:11:22 -0700720 } else if (vma->vm_end < pvma->vm_end)
David Howells8feae132009-01-08 12:04:47 +0000721 p = &(*p)->rb_left;
Namhyung Kim6038def2011-05-24 17:11:22 -0700722 else if (vma->vm_end > pvma->vm_end) {
723 rb_prev = parent;
David Howells8feae132009-01-08 12:04:47 +0000724 p = &(*p)->rb_right;
Namhyung Kim6038def2011-05-24 17:11:22 -0700725 } else if (vma < pvma)
David Howells8feae132009-01-08 12:04:47 +0000726 p = &(*p)->rb_left;
Namhyung Kim6038def2011-05-24 17:11:22 -0700727 else if (vma > pvma) {
728 rb_prev = parent;
David Howells8feae132009-01-08 12:04:47 +0000729 p = &(*p)->rb_right;
Namhyung Kim6038def2011-05-24 17:11:22 -0700730 } else
David Howells8feae132009-01-08 12:04:47 +0000731 BUG();
732 }
733
734 rb_link_node(&vma->vm_rb, parent, p);
735 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
736
737 /* add VMA to the VMA list also */
Namhyung Kim6038def2011-05-24 17:11:22 -0700738 prev = NULL;
739 if (rb_prev)
740 prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
David Howells30340972006-09-27 01:50:20 -0700741
Namhyung Kim6038def2011-05-24 17:11:22 -0700742 __vma_link_list(mm, vma, prev, parent);
David Howells8feae132009-01-08 12:04:47 +0000743}
744
745/*
746 * delete a VMA from its owning mm_struct and address space
747 */
748static void delete_vma_from_mm(struct vm_area_struct *vma)
749{
David Howells8feae132009-01-08 12:04:47 +0000750 struct address_space *mapping;
751 struct mm_struct *mm = vma->vm_mm;
752
753 kenter("%p", vma);
754
Bernd Schmidteb8cdec2009-09-21 17:03:57 -0700755 protect_vma(vma, 0);
756
David Howells8feae132009-01-08 12:04:47 +0000757 mm->map_count--;
758 if (mm->mmap_cache == vma)
759 mm->mmap_cache = NULL;
760
761 /* remove the VMA from the mapping */
762 if (vma->vm_file) {
763 mapping = vma->vm_file->f_mapping;
764
David Howells918e5562012-02-23 13:50:35 +0000765 mutex_lock(&mapping->i_mmap_mutex);
David Howells8feae132009-01-08 12:04:47 +0000766 flush_dcache_mmap_lock(mapping);
767 vma_prio_tree_remove(vma, &mapping->i_mmap);
768 flush_dcache_mmap_unlock(mapping);
David Howells918e5562012-02-23 13:50:35 +0000769 mutex_unlock(&mapping->i_mmap_mutex);
David Howells8feae132009-01-08 12:04:47 +0000770 }
771
772 /* remove from the MM's tree and list */
773 rb_erase(&vma->vm_rb, &mm->mm_rb);
Namhyung Kimb951bf22011-05-24 17:11:23 -0700774
775 if (vma->vm_prev)
776 vma->vm_prev->vm_next = vma->vm_next;
777 else
778 mm->mmap = vma->vm_next;
779
780 if (vma->vm_next)
781 vma->vm_next->vm_prev = vma->vm_prev;
David Howells8feae132009-01-08 12:04:47 +0000782}
783
784/*
785 * destroy a VMA record
786 */
787static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
788{
789 kenter("%p", vma);
790 if (vma->vm_ops && vma->vm_ops->close)
791 vma->vm_ops->close(vma);
792 if (vma->vm_file) {
793 fput(vma->vm_file);
794 if (vma->vm_flags & VM_EXECUTABLE)
795 removed_exe_file_vma(mm);
796 }
797 put_nommu_region(vma->vm_region);
798 kmem_cache_free(vm_area_cachep, vma);
David Howells30340972006-09-27 01:50:20 -0700799}
800
801/*
802 * look up the first VMA in which addr resides, NULL if none
803 * - should be called with mm->mmap_sem at least held readlocked
804 */
805struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
806{
David Howells8feae132009-01-08 12:04:47 +0000807 struct vm_area_struct *vma;
David Howells30340972006-09-27 01:50:20 -0700808
David Howells8feae132009-01-08 12:04:47 +0000809 /* check the cache first */
810 vma = mm->mmap_cache;
811 if (vma && vma->vm_start <= addr && vma->vm_end > addr)
812 return vma;
813
Namhyung Kime922c4c2011-05-24 17:11:24 -0700814 /* trawl the list (there may be multiple mappings in which addr
David Howells8feae132009-01-08 12:04:47 +0000815 * resides) */
Namhyung Kime922c4c2011-05-24 17:11:24 -0700816 for (vma = mm->mmap; vma; vma = vma->vm_next) {
David Howells8feae132009-01-08 12:04:47 +0000817 if (vma->vm_start > addr)
818 return NULL;
819 if (vma->vm_end > addr) {
820 mm->mmap_cache = vma;
821 return vma;
822 }
David Howells30340972006-09-27 01:50:20 -0700823 }
824
David Howells30340972006-09-27 01:50:20 -0700825 return NULL;
826}
827EXPORT_SYMBOL(find_vma);
828
829/*
David Howells930e6522006-09-27 01:50:22 -0700830 * find a VMA
831 * - we don't extend stack VMAs under NOMMU conditions
832 */
833struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
834{
David Howells7561e8c2010-03-25 16:48:38 +0000835 return find_vma(mm, addr);
David Howells930e6522006-09-27 01:50:22 -0700836}
837
David Howells8feae132009-01-08 12:04:47 +0000838/*
839 * expand a stack to a given address
840 * - not supported under NOMMU conditions
841 */
Greg Ungerer57c8f632007-07-15 23:38:28 -0700842int expand_stack(struct vm_area_struct *vma, unsigned long address)
843{
844 return -ENOMEM;
845}
846
David Howells930e6522006-09-27 01:50:22 -0700847/*
David Howells6fa5f802006-09-27 01:50:21 -0700848 * look up the first VMA exactly that exactly matches addr
849 * - should be called with mm->mmap_sem at least held readlocked
850 */
David Howells8feae132009-01-08 12:04:47 +0000851static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
852 unsigned long addr,
853 unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854{
855 struct vm_area_struct *vma;
David Howells8feae132009-01-08 12:04:47 +0000856 unsigned long end = addr + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857
David Howells8feae132009-01-08 12:04:47 +0000858 /* check the cache first */
859 vma = mm->mmap_cache;
860 if (vma && vma->vm_start == addr && vma->vm_end == end)
861 return vma;
862
Namhyung Kime922c4c2011-05-24 17:11:24 -0700863 /* trawl the list (there may be multiple mappings in which addr
David Howells8feae132009-01-08 12:04:47 +0000864 * resides) */
Namhyung Kime922c4c2011-05-24 17:11:24 -0700865 for (vma = mm->mmap; vma; vma = vma->vm_next) {
David Howells8feae132009-01-08 12:04:47 +0000866 if (vma->vm_start < addr)
867 continue;
868 if (vma->vm_start > addr)
869 return NULL;
870 if (vma->vm_end == end) {
871 mm->mmap_cache = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 return vma;
David Howells8feae132009-01-08 12:04:47 +0000873 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 }
875
876 return NULL;
877}
878
David Howells30340972006-09-27 01:50:20 -0700879/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 * determine whether a mapping should be permitted and, if so, what sort of
881 * mapping we're capable of supporting
882 */
883static int validate_mmap_request(struct file *file,
884 unsigned long addr,
885 unsigned long len,
886 unsigned long prot,
887 unsigned long flags,
888 unsigned long pgoff,
889 unsigned long *_capabilities)
890{
David Howells8feae132009-01-08 12:04:47 +0000891 unsigned long capabilities, rlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 unsigned long reqprot = prot;
893 int ret;
894
895 /* do the simple checks first */
David Howells06aab5a2009-09-24 12:33:48 +0100896 if (flags & MAP_FIXED) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 printk(KERN_DEBUG
898 "%d: Can't do fixed-address/overlay mmap of RAM\n",
899 current->pid);
900 return -EINVAL;
901 }
902
903 if ((flags & MAP_TYPE) != MAP_PRIVATE &&
904 (flags & MAP_TYPE) != MAP_SHARED)
905 return -EINVAL;
906
Mike Frysingerf81cff02006-12-06 12:02:59 +1000907 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 return -EINVAL;
909
Mike Frysingerf81cff02006-12-06 12:02:59 +1000910 /* Careful about overflows.. */
David Howells8feae132009-01-08 12:04:47 +0000911 rlen = PAGE_ALIGN(len);
912 if (!rlen || rlen > TASK_SIZE)
Mike Frysingerf81cff02006-12-06 12:02:59 +1000913 return -ENOMEM;
914
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 /* offset overflow? */
David Howells8feae132009-01-08 12:04:47 +0000916 if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
Mike Frysingerf81cff02006-12-06 12:02:59 +1000917 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918
919 if (file) {
920 /* validate file mapping requests */
921 struct address_space *mapping;
922
923 /* files must support mmap */
924 if (!file->f_op || !file->f_op->mmap)
925 return -ENODEV;
926
927 /* work out if what we've got could possibly be shared
928 * - we support chardevs that provide their own "memory"
929 * - we support files/blockdevs that are memory backed
930 */
931 mapping = file->f_mapping;
932 if (!mapping)
Josef Sipeke9536ae2006-12-08 02:37:21 -0800933 mapping = file->f_path.dentry->d_inode->i_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934
935 capabilities = 0;
936 if (mapping && mapping->backing_dev_info)
937 capabilities = mapping->backing_dev_info->capabilities;
938
939 if (!capabilities) {
940 /* no explicit capabilities set, so assume some
941 * defaults */
Josef Sipeke9536ae2006-12-08 02:37:21 -0800942 switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943 case S_IFREG:
944 case S_IFBLK:
945 capabilities = BDI_CAP_MAP_COPY;
946 break;
947
948 case S_IFCHR:
949 capabilities =
950 BDI_CAP_MAP_DIRECT |
951 BDI_CAP_READ_MAP |
952 BDI_CAP_WRITE_MAP;
953 break;
954
955 default:
956 return -EINVAL;
957 }
958 }
959
960 /* eliminate any capabilities that we can't support on this
961 * device */
962 if (!file->f_op->get_unmapped_area)
963 capabilities &= ~BDI_CAP_MAP_DIRECT;
964 if (!file->f_op->read)
965 capabilities &= ~BDI_CAP_MAP_COPY;
966
Graff Yang28d7a6a2009-08-18 14:11:17 -0700967 /* The file shall have been opened with read permission. */
968 if (!(file->f_mode & FMODE_READ))
969 return -EACCES;
970
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 if (flags & MAP_SHARED) {
972 /* do checks for writing, appending and locking */
973 if ((prot & PROT_WRITE) &&
974 !(file->f_mode & FMODE_WRITE))
975 return -EACCES;
976
Josef Sipeke9536ae2006-12-08 02:37:21 -0800977 if (IS_APPEND(file->f_path.dentry->d_inode) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 (file->f_mode & FMODE_WRITE))
979 return -EACCES;
980
Josef Sipeke9536ae2006-12-08 02:37:21 -0800981 if (locks_verify_locked(file->f_path.dentry->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 return -EAGAIN;
983
984 if (!(capabilities & BDI_CAP_MAP_DIRECT))
985 return -ENODEV;
986
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 /* we mustn't privatise shared mappings */
988 capabilities &= ~BDI_CAP_MAP_COPY;
989 }
990 else {
991 /* we're going to read the file into private memory we
992 * allocate */
993 if (!(capabilities & BDI_CAP_MAP_COPY))
994 return -ENODEV;
995
996 /* we don't permit a private writable mapping to be
997 * shared with the backing device */
998 if (prot & PROT_WRITE)
999 capabilities &= ~BDI_CAP_MAP_DIRECT;
1000 }
1001
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001002 if (capabilities & BDI_CAP_MAP_DIRECT) {
1003 if (((prot & PROT_READ) && !(capabilities & BDI_CAP_READ_MAP)) ||
1004 ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
1005 ((prot & PROT_EXEC) && !(capabilities & BDI_CAP_EXEC_MAP))
1006 ) {
1007 capabilities &= ~BDI_CAP_MAP_DIRECT;
1008 if (flags & MAP_SHARED) {
1009 printk(KERN_WARNING
1010 "MAP_SHARED not completely supported on !MMU\n");
1011 return -EINVAL;
1012 }
1013 }
1014 }
1015
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016 /* handle executable mappings and implied executable
1017 * mappings */
Josef Sipeke9536ae2006-12-08 02:37:21 -08001018 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 if (prot & PROT_EXEC)
1020 return -EPERM;
1021 }
1022 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
1023 /* handle implication of PROT_EXEC by PROT_READ */
1024 if (current->personality & READ_IMPLIES_EXEC) {
1025 if (capabilities & BDI_CAP_EXEC_MAP)
1026 prot |= PROT_EXEC;
1027 }
1028 }
1029 else if ((prot & PROT_READ) &&
1030 (prot & PROT_EXEC) &&
1031 !(capabilities & BDI_CAP_EXEC_MAP)
1032 ) {
1033 /* backing file is not executable, try to copy */
1034 capabilities &= ~BDI_CAP_MAP_DIRECT;
1035 }
1036 }
1037 else {
1038 /* anonymous mappings are always memory backed and can be
1039 * privately mapped
1040 */
1041 capabilities = BDI_CAP_MAP_COPY;
1042
1043 /* handle PROT_EXEC implication by PROT_READ */
1044 if ((prot & PROT_READ) &&
1045 (current->personality & READ_IMPLIES_EXEC))
1046 prot |= PROT_EXEC;
1047 }
1048
1049 /* allow the security API to have its say */
Eric Parised032182007-06-28 15:55:21 -04001050 ret = security_file_mmap(file, reqprot, prot, flags, addr, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 if (ret < 0)
1052 return ret;
1053
1054 /* looks okay */
1055 *_capabilities = capabilities;
1056 return 0;
1057}
1058
1059/*
1060 * we've determined that we can make the mapping, now translate what we
1061 * now know into VMA flags
1062 */
1063static unsigned long determine_vm_flags(struct file *file,
1064 unsigned long prot,
1065 unsigned long flags,
1066 unsigned long capabilities)
1067{
1068 unsigned long vm_flags;
1069
1070 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071 /* vm_flags |= mm->def_flags; */
1072
1073 if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
1074 /* attempt to share read-only copies of mapped file chunks */
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001075 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 if (file && !(prot & PROT_WRITE))
1077 vm_flags |= VM_MAYSHARE;
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001078 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 /* overlay a shareable mapping on the backing device or inode
1080 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1081 * romfs/cramfs */
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001082 vm_flags |= VM_MAYSHARE | (capabilities & BDI_CAP_VMFLAGS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 if (flags & MAP_SHARED)
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001084 vm_flags |= VM_SHARED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 }
1086
1087 /* refuse to let anyone share private mappings with this process if
1088 * it's being traced - otherwise breakpoints set in it may interfere
1089 * with another untraced process
1090 */
Tejun Heoa288eec2011-06-17 16:50:37 +02001091 if ((flags & MAP_PRIVATE) && current->ptrace)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 vm_flags &= ~VM_MAYSHARE;
1093
1094 return vm_flags;
1095}
1096
1097/*
David Howells8feae132009-01-08 12:04:47 +00001098 * set up a shared mapping on a file (the driver or filesystem provides and
1099 * pins the storage)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 */
David Howells8feae132009-01-08 12:04:47 +00001101static int do_mmap_shared_file(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102{
1103 int ret;
1104
1105 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001106 if (ret == 0) {
1107 vma->vm_region->vm_top = vma->vm_region->vm_end;
David Howells645d83c2009-09-24 15:13:10 +01001108 return 0;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001109 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 if (ret != -ENOSYS)
1111 return ret;
1112
David Howells3fa30462010-03-23 13:35:21 -07001113 /* getting -ENOSYS indicates that direct mmap isn't possible (as
1114 * opposed to tried but failed) so we can only give a suitable error as
1115 * it's not possible to make a private copy if MAP_SHARED was given */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 return -ENODEV;
1117}
1118
1119/*
1120 * set up a private mapping or an anonymous shared mapping
1121 */
David Howells8feae132009-01-08 12:04:47 +00001122static int do_mmap_private(struct vm_area_struct *vma,
1123 struct vm_region *region,
David Howells645d83c2009-09-24 15:13:10 +01001124 unsigned long len,
1125 unsigned long capabilities)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126{
David Howells8feae132009-01-08 12:04:47 +00001127 struct page *pages;
Bob Liuf67d9b12011-05-24 17:12:56 -07001128 unsigned long total, point, n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 void *base;
David Howells8feae132009-01-08 12:04:47 +00001130 int ret, order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131
1132 /* invoke the file's mapping function so that it can keep track of
1133 * shared mappings on devices or memory
1134 * - VM_MAYSHARE will be set if it may attempt to share
1135 */
David Howells645d83c2009-09-24 15:13:10 +01001136 if (capabilities & BDI_CAP_MAP_DIRECT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001138 if (ret == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139 /* shouldn't return success if we're not sharing */
Paul Mundtdd8632a2009-01-08 12:04:47 +00001140 BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1141 vma->vm_region->vm_top = vma->vm_region->vm_end;
David Howells645d83c2009-09-24 15:13:10 +01001142 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 }
Paul Mundtdd8632a2009-01-08 12:04:47 +00001144 if (ret != -ENOSYS)
1145 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146
1147 /* getting an ENOSYS error indicates that direct mmap isn't
1148 * possible (as opposed to tried but failed) so we'll try to
1149 * make a private copy of the data and map that instead */
1150 }
1151
David Howells8feae132009-01-08 12:04:47 +00001152
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 /* allocate some memory to hold the mapping
1154 * - note that this may not return a page-aligned address if the object
1155 * we're allocating is smaller than a page
1156 */
Bob Liuf67d9b12011-05-24 17:12:56 -07001157 order = get_order(len);
David Howells8feae132009-01-08 12:04:47 +00001158 kdebug("alloc order %d for %lx", order, len);
1159
1160 pages = alloc_pages(GFP_KERNEL, order);
1161 if (!pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162 goto enomem;
1163
David Howells8feae132009-01-08 12:04:47 +00001164 total = 1 << order;
David Howells33e5d7692009-04-02 16:56:32 -07001165 atomic_long_add(total, &mmap_pages_allocated);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166
Bob Liuf67d9b12011-05-24 17:12:56 -07001167 point = len >> PAGE_SHIFT;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001168
1169 /* we allocated a power-of-2 sized page set, so we may want to trim off
1170 * the excess */
1171 if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) {
1172 while (total > point) {
1173 order = ilog2(total - point);
1174 n = 1 << order;
1175 kdebug("shave %lu/%lu @%lu", n, total - point, total);
David Howells33e5d7692009-04-02 16:56:32 -07001176 atomic_long_sub(n, &mmap_pages_allocated);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001177 total -= n;
1178 set_page_refcounted(pages + total);
1179 __free_pages(pages + total, order);
1180 }
David Howells8feae132009-01-08 12:04:47 +00001181 }
1182
David Howells8feae132009-01-08 12:04:47 +00001183 for (point = 1; point < total; point++)
1184 set_page_refcounted(&pages[point]);
1185
1186 base = page_address(pages);
1187 region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1188 region->vm_start = (unsigned long) base;
Bob Liuf67d9b12011-05-24 17:12:56 -07001189 region->vm_end = region->vm_start + len;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001190 region->vm_top = region->vm_start + (total << PAGE_SHIFT);
David Howells8feae132009-01-08 12:04:47 +00001191
1192 vma->vm_start = region->vm_start;
1193 vma->vm_end = region->vm_start + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194
1195 if (vma->vm_file) {
1196 /* read the contents of a file into the copy */
1197 mm_segment_t old_fs;
1198 loff_t fpos;
1199
1200 fpos = vma->vm_pgoff;
1201 fpos <<= PAGE_SHIFT;
1202
1203 old_fs = get_fs();
1204 set_fs(KERNEL_DS);
Bob Liuf67d9b12011-05-24 17:12:56 -07001205 ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 set_fs(old_fs);
1207
1208 if (ret < 0)
1209 goto error_free;
1210
1211 /* clear the last little bit */
Bob Liuf67d9b12011-05-24 17:12:56 -07001212 if (ret < len)
1213 memset(base + ret, 0, len - ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 }
1216
1217 return 0;
1218
1219error_free:
Namhyung Kim7223bb42011-05-24 17:11:26 -07001220 free_page_series(region->vm_start, region->vm_top);
David Howells8feae132009-01-08 12:04:47 +00001221 region->vm_start = vma->vm_start = 0;
1222 region->vm_end = vma->vm_end = 0;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001223 region->vm_top = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 return ret;
1225
1226enomem:
Greg Ungerer05ae6fa2009-01-13 17:30:22 +10001227 printk("Allocation of length %lu from process %d (%s) failed\n",
1228 len, current->pid, current->comm);
David Rientjes7bf02ea2011-05-24 17:11:16 -07001229 show_free_areas(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 return -ENOMEM;
1231}
1232
1233/*
1234 * handle mapping creation for uClinux
1235 */
1236unsigned long do_mmap_pgoff(struct file *file,
1237 unsigned long addr,
1238 unsigned long len,
1239 unsigned long prot,
1240 unsigned long flags,
1241 unsigned long pgoff)
1242{
David Howells8feae132009-01-08 12:04:47 +00001243 struct vm_area_struct *vma;
1244 struct vm_region *region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 struct rb_node *rb;
David Howells8feae132009-01-08 12:04:47 +00001246 unsigned long capabilities, vm_flags, result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247 int ret;
1248
David Howells8feae132009-01-08 12:04:47 +00001249 kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff);
1250
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251 /* decide whether we should attempt the mapping, and if so what sort of
1252 * mapping */
1253 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1254 &capabilities);
David Howells8feae132009-01-08 12:04:47 +00001255 if (ret < 0) {
1256 kleave(" = %d [val]", ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 return ret;
David Howells8feae132009-01-08 12:04:47 +00001258 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259
David Howells06aab5a2009-09-24 12:33:48 +01001260 /* we ignore the address hint */
1261 addr = 0;
Bob Liuf67d9b12011-05-24 17:12:56 -07001262 len = PAGE_ALIGN(len);
David Howells06aab5a2009-09-24 12:33:48 +01001263
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 /* we've determined that we can make the mapping, now translate what we
1265 * now know into VMA flags */
1266 vm_flags = determine_vm_flags(file, prot, flags, capabilities);
1267
David Howells8feae132009-01-08 12:04:47 +00001268 /* we're going to need to record the mapping */
1269 region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1270 if (!region)
1271 goto error_getting_region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272
David Howells8feae132009-01-08 12:04:47 +00001273 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1274 if (!vma)
1275 goto error_getting_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276
David Howells1e2ae592010-01-15 17:01:33 -08001277 region->vm_usage = 1;
David Howells8feae132009-01-08 12:04:47 +00001278 region->vm_flags = vm_flags;
1279 region->vm_pgoff = pgoff;
1280
Rik van Riel5beb4932010-03-05 13:42:07 -08001281 INIT_LIST_HEAD(&vma->anon_vma_chain);
David Howells8feae132009-01-08 12:04:47 +00001282 vma->vm_flags = vm_flags;
1283 vma->vm_pgoff = pgoff;
1284
1285 if (file) {
1286 region->vm_file = file;
1287 get_file(file);
1288 vma->vm_file = file;
1289 get_file(file);
1290 if (vm_flags & VM_EXECUTABLE) {
1291 added_exe_file_vma(current->mm);
1292 vma->vm_mm = current->mm;
1293 }
1294 }
1295
1296 down_write(&nommu_region_sem);
1297
1298 /* if we want to share, we need to check for regions created by other
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 * mmap() calls that overlap with our proposed mapping
David Howells8feae132009-01-08 12:04:47 +00001300 * - we can only share with a superset match on most regular files
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301 * - shared mappings on character devices and memory backed files are
1302 * permitted to overlap inexactly as far as we are concerned for in
1303 * these cases, sharing is handled in the driver or filesystem rather
1304 * than here
1305 */
1306 if (vm_flags & VM_MAYSHARE) {
David Howells8feae132009-01-08 12:04:47 +00001307 struct vm_region *pregion;
1308 unsigned long pglen, rpglen, pgend, rpgend, start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309
David Howells8feae132009-01-08 12:04:47 +00001310 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1311 pgend = pgoff + pglen;
David Howells165b2392007-03-22 00:11:24 -08001312
David Howells8feae132009-01-08 12:04:47 +00001313 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1314 pregion = rb_entry(rb, struct vm_region, vm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315
David Howells8feae132009-01-08 12:04:47 +00001316 if (!(pregion->vm_flags & VM_MAYSHARE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317 continue;
1318
1319 /* search for overlapping mappings on the same file */
David Howells8feae132009-01-08 12:04:47 +00001320 if (pregion->vm_file->f_path.dentry->d_inode !=
1321 file->f_path.dentry->d_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322 continue;
1323
David Howells8feae132009-01-08 12:04:47 +00001324 if (pregion->vm_pgoff >= pgend)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 continue;
1326
David Howells8feae132009-01-08 12:04:47 +00001327 rpglen = pregion->vm_end - pregion->vm_start;
1328 rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1329 rpgend = pregion->vm_pgoff + rpglen;
1330 if (pgoff >= rpgend)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331 continue;
1332
David Howells8feae132009-01-08 12:04:47 +00001333 /* handle inexactly overlapping matches between
1334 * mappings */
1335 if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1336 !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1337 /* new mapping is not a subset of the region */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 if (!(capabilities & BDI_CAP_MAP_DIRECT))
1339 goto sharing_violation;
1340 continue;
1341 }
1342
David Howells8feae132009-01-08 12:04:47 +00001343 /* we've found a region we can share */
David Howells1e2ae592010-01-15 17:01:33 -08001344 pregion->vm_usage++;
David Howells8feae132009-01-08 12:04:47 +00001345 vma->vm_region = pregion;
1346 start = pregion->vm_start;
1347 start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1348 vma->vm_start = start;
1349 vma->vm_end = start + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350
David Howells8feae132009-01-08 12:04:47 +00001351 if (pregion->vm_flags & VM_MAPPED_COPY) {
1352 kdebug("share copy");
1353 vma->vm_flags |= VM_MAPPED_COPY;
1354 } else {
1355 kdebug("share mmap");
1356 ret = do_mmap_shared_file(vma);
1357 if (ret < 0) {
1358 vma->vm_region = NULL;
1359 vma->vm_start = 0;
1360 vma->vm_end = 0;
David Howells1e2ae592010-01-15 17:01:33 -08001361 pregion->vm_usage--;
David Howells8feae132009-01-08 12:04:47 +00001362 pregion = NULL;
1363 goto error_just_free;
1364 }
1365 }
1366 fput(region->vm_file);
1367 kmem_cache_free(vm_region_jar, region);
1368 region = pregion;
1369 result = start;
1370 goto share;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 }
1372
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 /* obtain the address at which to make a shared mapping
1374 * - this is the hook for quasi-memory character devices to
1375 * tell us the location of a shared mapping
1376 */
David Howells645d83c2009-09-24 15:13:10 +01001377 if (capabilities & BDI_CAP_MAP_DIRECT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378 addr = file->f_op->get_unmapped_area(file, addr, len,
1379 pgoff, flags);
Namhyung Kimbb005a52011-05-24 17:11:27 -07001380 if (IS_ERR_VALUE(addr)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 ret = addr;
Namhyung Kimbb005a52011-05-24 17:11:27 -07001382 if (ret != -ENOSYS)
David Howells8feae132009-01-08 12:04:47 +00001383 goto error_just_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384
1385 /* the driver refused to tell us where to site
1386 * the mapping so we'll have to attempt to copy
1387 * it */
Namhyung Kimbb005a52011-05-24 17:11:27 -07001388 ret = -ENODEV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389 if (!(capabilities & BDI_CAP_MAP_COPY))
David Howells8feae132009-01-08 12:04:47 +00001390 goto error_just_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391
1392 capabilities &= ~BDI_CAP_MAP_DIRECT;
David Howells8feae132009-01-08 12:04:47 +00001393 } else {
1394 vma->vm_start = region->vm_start = addr;
1395 vma->vm_end = region->vm_end = addr + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396 }
1397 }
1398 }
1399
David Howells8feae132009-01-08 12:04:47 +00001400 vma->vm_region = region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401
David Howells645d83c2009-09-24 15:13:10 +01001402 /* set up the mapping
1403 * - the region is filled in if BDI_CAP_MAP_DIRECT is still set
1404 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405 if (file && vma->vm_flags & VM_SHARED)
David Howells8feae132009-01-08 12:04:47 +00001406 ret = do_mmap_shared_file(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407 else
David Howells645d83c2009-09-24 15:13:10 +01001408 ret = do_mmap_private(vma, region, len, capabilities);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 if (ret < 0)
David Howells645d83c2009-09-24 15:13:10 +01001410 goto error_just_free;
1411 add_nommu_region(region);
David Howells8feae132009-01-08 12:04:47 +00001412
Jie Zhangea637632009-12-14 18:00:02 -08001413 /* clear anonymous mappings that don't ask for uninitialized data */
1414 if (!vma->vm_file && !(flags & MAP_UNINITIALIZED))
1415 memset((void *)region->vm_start, 0,
1416 region->vm_end - region->vm_start);
1417
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418 /* okay... we have a mapping; now we have to register it */
David Howells8feae132009-01-08 12:04:47 +00001419 result = vma->vm_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 current->mm->total_vm += len >> PAGE_SHIFT;
1422
David Howells8feae132009-01-08 12:04:47 +00001423share:
1424 add_vma_to_mm(current->mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425
Mike Frysingercfe79c02010-01-06 17:23:23 +00001426 /* we flush the region from the icache only when the first executable
1427 * mapping of it is made */
1428 if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1429 flush_icache_range(region->vm_start, region->vm_end);
1430 region->vm_icache_flushed = true;
1431 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432
Mike Frysingercfe79c02010-01-06 17:23:23 +00001433 up_write(&nommu_region_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434
David Howells8feae132009-01-08 12:04:47 +00001435 kleave(" = %lx", result);
1436 return result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437
David Howells8feae132009-01-08 12:04:47 +00001438error_just_free:
1439 up_write(&nommu_region_sem);
1440error:
David Howells89a86402009-10-30 13:13:26 +00001441 if (region->vm_file)
1442 fput(region->vm_file);
David Howells8feae132009-01-08 12:04:47 +00001443 kmem_cache_free(vm_region_jar, region);
David Howells89a86402009-10-30 13:13:26 +00001444 if (vma->vm_file)
1445 fput(vma->vm_file);
David Howells8feae132009-01-08 12:04:47 +00001446 if (vma->vm_flags & VM_EXECUTABLE)
1447 removed_exe_file_vma(vma->vm_mm);
1448 kmem_cache_free(vm_area_cachep, vma);
1449 kleave(" = %d", ret);
1450 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451
David Howells8feae132009-01-08 12:04:47 +00001452sharing_violation:
1453 up_write(&nommu_region_sem);
1454 printk(KERN_WARNING "Attempt to share mismatched mappings\n");
1455 ret = -EINVAL;
1456 goto error;
1457
1458error_getting_vma:
1459 kmem_cache_free(vm_region_jar, region);
1460 printk(KERN_WARNING "Allocation of vma for %lu byte allocation"
1461 " from process %d failed\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 len, current->pid);
David Rientjes7bf02ea2011-05-24 17:11:16 -07001463 show_free_areas(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464 return -ENOMEM;
1465
David Howells8feae132009-01-08 12:04:47 +00001466error_getting_region:
1467 printk(KERN_WARNING "Allocation of vm region for %lu byte allocation"
1468 " from process %d failed\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 len, current->pid);
David Rientjes7bf02ea2011-05-24 17:11:16 -07001470 show_free_areas(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 return -ENOMEM;
1472}
Paul Mundtb5073172007-07-21 04:37:25 -07001473EXPORT_SYMBOL(do_mmap_pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001475SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1476 unsigned long, prot, unsigned long, flags,
1477 unsigned long, fd, unsigned long, pgoff)
1478{
1479 struct file *file = NULL;
1480 unsigned long retval = -EBADF;
1481
Al Viro120a7952010-10-30 02:54:44 -04001482 audit_mmap_fd(fd, flags);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001483 if (!(flags & MAP_ANONYMOUS)) {
1484 file = fget(fd);
1485 if (!file)
1486 goto out;
1487 }
1488
1489 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1490
1491 down_write(&current->mm->mmap_sem);
1492 retval = do_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1493 up_write(&current->mm->mmap_sem);
1494
1495 if (file)
1496 fput(file);
1497out:
1498 return retval;
1499}
1500
Christoph Hellwiga4679372010-03-10 15:21:15 -08001501#ifdef __ARCH_WANT_SYS_OLD_MMAP
1502struct mmap_arg_struct {
1503 unsigned long addr;
1504 unsigned long len;
1505 unsigned long prot;
1506 unsigned long flags;
1507 unsigned long fd;
1508 unsigned long offset;
1509};
1510
1511SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1512{
1513 struct mmap_arg_struct a;
1514
1515 if (copy_from_user(&a, arg, sizeof(a)))
1516 return -EFAULT;
1517 if (a.offset & ~PAGE_MASK)
1518 return -EINVAL;
1519
1520 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1521 a.offset >> PAGE_SHIFT);
1522}
1523#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1524
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525/*
David Howells8feae132009-01-08 12:04:47 +00001526 * split a vma into two pieces at address 'addr', a new vma is allocated either
1527 * for the first part or the tail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 */
David Howells8feae132009-01-08 12:04:47 +00001529int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1530 unsigned long addr, int new_below)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531{
David Howells8feae132009-01-08 12:04:47 +00001532 struct vm_area_struct *new;
1533 struct vm_region *region;
1534 unsigned long npages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535
David Howells8feae132009-01-08 12:04:47 +00001536 kenter("");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537
David Howells779c1022010-01-15 17:01:34 -08001538 /* we're only permitted to split anonymous regions (these should have
1539 * only a single usage on the region) */
1540 if (vma->vm_file)
David Howells8feae132009-01-08 12:04:47 +00001541 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542
David Howells8feae132009-01-08 12:04:47 +00001543 if (mm->map_count >= sysctl_max_map_count)
1544 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545
David Howells8feae132009-01-08 12:04:47 +00001546 region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1547 if (!region)
1548 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549
David Howells8feae132009-01-08 12:04:47 +00001550 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1551 if (!new) {
1552 kmem_cache_free(vm_region_jar, region);
1553 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554 }
David Howells8feae132009-01-08 12:04:47 +00001555
1556 /* most fields are the same, copy all, and then fixup */
1557 *new = *vma;
1558 *region = *vma->vm_region;
1559 new->vm_region = region;
1560
1561 npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1562
1563 if (new_below) {
Paul Mundtdd8632a2009-01-08 12:04:47 +00001564 region->vm_top = region->vm_end = new->vm_end = addr;
David Howells8feae132009-01-08 12:04:47 +00001565 } else {
1566 region->vm_start = new->vm_start = addr;
1567 region->vm_pgoff = new->vm_pgoff += npages;
1568 }
1569
1570 if (new->vm_ops && new->vm_ops->open)
1571 new->vm_ops->open(new);
1572
1573 delete_vma_from_mm(vma);
1574 down_write(&nommu_region_sem);
1575 delete_nommu_region(vma->vm_region);
1576 if (new_below) {
1577 vma->vm_region->vm_start = vma->vm_start = addr;
1578 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1579 } else {
1580 vma->vm_region->vm_end = vma->vm_end = addr;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001581 vma->vm_region->vm_top = addr;
David Howells8feae132009-01-08 12:04:47 +00001582 }
1583 add_nommu_region(vma->vm_region);
1584 add_nommu_region(new->vm_region);
1585 up_write(&nommu_region_sem);
1586 add_vma_to_mm(mm, vma);
1587 add_vma_to_mm(mm, new);
1588 return 0;
1589}
1590
1591/*
1592 * shrink a VMA by removing the specified chunk from either the beginning or
1593 * the end
1594 */
1595static int shrink_vma(struct mm_struct *mm,
1596 struct vm_area_struct *vma,
1597 unsigned long from, unsigned long to)
1598{
1599 struct vm_region *region;
1600
1601 kenter("");
1602
1603 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1604 * and list */
1605 delete_vma_from_mm(vma);
1606 if (from > vma->vm_start)
1607 vma->vm_end = from;
1608 else
1609 vma->vm_start = to;
1610 add_vma_to_mm(mm, vma);
1611
1612 /* cut the backing region down to size */
1613 region = vma->vm_region;
David Howells1e2ae592010-01-15 17:01:33 -08001614 BUG_ON(region->vm_usage != 1);
David Howells8feae132009-01-08 12:04:47 +00001615
1616 down_write(&nommu_region_sem);
1617 delete_nommu_region(region);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001618 if (from > region->vm_start) {
1619 to = region->vm_top;
1620 region->vm_top = region->vm_end = from;
1621 } else {
David Howells8feae132009-01-08 12:04:47 +00001622 region->vm_start = to;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001623 }
David Howells8feae132009-01-08 12:04:47 +00001624 add_nommu_region(region);
1625 up_write(&nommu_region_sem);
1626
1627 free_page_series(from, to);
1628 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629}
1630
David Howells30340972006-09-27 01:50:20 -07001631/*
1632 * release a mapping
David Howells8feae132009-01-08 12:04:47 +00001633 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1634 * VMA, though it need not cover the whole VMA
David Howells30340972006-09-27 01:50:20 -07001635 */
David Howells8feae132009-01-08 12:04:47 +00001636int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637{
David Howells8feae132009-01-08 12:04:47 +00001638 struct vm_area_struct *vma;
Bob Liuf67d9b12011-05-24 17:12:56 -07001639 unsigned long end;
David Howells8feae132009-01-08 12:04:47 +00001640 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641
David Howells8feae132009-01-08 12:04:47 +00001642 kenter(",%lx,%zx", start, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643
Bob Liuf67d9b12011-05-24 17:12:56 -07001644 len = PAGE_ALIGN(len);
David Howells8feae132009-01-08 12:04:47 +00001645 if (len == 0)
1646 return -EINVAL;
1647
Bob Liuf67d9b12011-05-24 17:12:56 -07001648 end = start + len;
1649
David Howells8feae132009-01-08 12:04:47 +00001650 /* find the first potentially overlapping VMA */
1651 vma = find_vma(mm, start);
1652 if (!vma) {
David Howells33e5d7692009-04-02 16:56:32 -07001653 static int limit = 0;
1654 if (limit < 5) {
1655 printk(KERN_WARNING
1656 "munmap of memory not mmapped by process %d"
1657 " (%s): 0x%lx-0x%lx\n",
1658 current->pid, current->comm,
1659 start, start + len - 1);
1660 limit++;
1661 }
David Howells8feae132009-01-08 12:04:47 +00001662 return -EINVAL;
David Howells30340972006-09-27 01:50:20 -07001663 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664
David Howells8feae132009-01-08 12:04:47 +00001665 /* we're allowed to split an anonymous VMA but not a file-backed one */
1666 if (vma->vm_file) {
1667 do {
1668 if (start > vma->vm_start) {
1669 kleave(" = -EINVAL [miss]");
1670 return -EINVAL;
1671 }
1672 if (end == vma->vm_end)
1673 goto erase_whole_vma;
Namhyung Kimd75a3102011-05-24 17:11:25 -07001674 vma = vma->vm_next;
1675 } while (vma);
David Howells8feae132009-01-08 12:04:47 +00001676 kleave(" = -EINVAL [split file]");
1677 return -EINVAL;
1678 } else {
1679 /* the chunk must be a subset of the VMA found */
1680 if (start == vma->vm_start && end == vma->vm_end)
1681 goto erase_whole_vma;
1682 if (start < vma->vm_start || end > vma->vm_end) {
1683 kleave(" = -EINVAL [superset]");
1684 return -EINVAL;
1685 }
1686 if (start & ~PAGE_MASK) {
1687 kleave(" = -EINVAL [unaligned start]");
1688 return -EINVAL;
1689 }
1690 if (end != vma->vm_end && end & ~PAGE_MASK) {
1691 kleave(" = -EINVAL [unaligned split]");
1692 return -EINVAL;
1693 }
1694 if (start != vma->vm_start && end != vma->vm_end) {
1695 ret = split_vma(mm, vma, start, 1);
1696 if (ret < 0) {
1697 kleave(" = %d [split]", ret);
1698 return ret;
1699 }
1700 }
1701 return shrink_vma(mm, vma, start, end);
1702 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703
David Howells8feae132009-01-08 12:04:47 +00001704erase_whole_vma:
1705 delete_vma_from_mm(vma);
1706 delete_vma(mm, vma);
1707 kleave(" = 0");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708 return 0;
1709}
Paul Mundtb5073172007-07-21 04:37:25 -07001710EXPORT_SYMBOL(do_munmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711
Heiko Carstens6a6160a2009-01-14 14:14:15 +01001712SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
David Howells30340972006-09-27 01:50:20 -07001713{
1714 int ret;
1715 struct mm_struct *mm = current->mm;
1716
1717 down_write(&mm->mmap_sem);
1718 ret = do_munmap(mm, addr, len);
1719 up_write(&mm->mmap_sem);
1720 return ret;
1721}
1722
1723/*
David Howells8feae132009-01-08 12:04:47 +00001724 * release all the mappings made in a process's VM space
David Howells30340972006-09-27 01:50:20 -07001725 */
David Howells8feae132009-01-08 12:04:47 +00001726void exit_mmap(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727{
David Howells8feae132009-01-08 12:04:47 +00001728 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729
David Howells8feae132009-01-08 12:04:47 +00001730 if (!mm)
1731 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732
David Howells8feae132009-01-08 12:04:47 +00001733 kenter("");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734
David Howells8feae132009-01-08 12:04:47 +00001735 mm->total_vm = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736
David Howells8feae132009-01-08 12:04:47 +00001737 while ((vma = mm->mmap)) {
1738 mm->mmap = vma->vm_next;
1739 delete_vma_from_mm(vma);
1740 delete_vma(mm, vma);
Steven J. Magnani04c34962010-11-24 12:56:54 -08001741 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742 }
David Howells8feae132009-01-08 12:04:47 +00001743
1744 kleave("");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745}
1746
Linus Torvaldse4eb1ff2012-04-20 15:35:40 -07001747unsigned long vm_brk(unsigned long addr, unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748{
1749 return -ENOMEM;
1750}
1751
1752/*
David Howells6fa5f802006-09-27 01:50:21 -07001753 * expand (or shrink) an existing mapping, potentially moving it at the same
1754 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 *
David Howells6fa5f802006-09-27 01:50:21 -07001756 * under NOMMU conditions, we only permit changing a mapping's size, and only
David Howells8feae132009-01-08 12:04:47 +00001757 * as long as it stays within the region allocated by do_mmap_private() and the
1758 * block is not shareable
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759 *
David Howells6fa5f802006-09-27 01:50:21 -07001760 * MREMAP_FIXED is not supported under NOMMU conditions
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761 */
1762unsigned long do_mremap(unsigned long addr,
1763 unsigned long old_len, unsigned long new_len,
1764 unsigned long flags, unsigned long new_addr)
1765{
David Howells6fa5f802006-09-27 01:50:21 -07001766 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767
1768 /* insanity checks first */
Bob Liuf67d9b12011-05-24 17:12:56 -07001769 old_len = PAGE_ALIGN(old_len);
1770 new_len = PAGE_ALIGN(new_len);
David Howells8feae132009-01-08 12:04:47 +00001771 if (old_len == 0 || new_len == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772 return (unsigned long) -EINVAL;
1773
David Howells8feae132009-01-08 12:04:47 +00001774 if (addr & ~PAGE_MASK)
1775 return -EINVAL;
1776
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777 if (flags & MREMAP_FIXED && new_addr != addr)
1778 return (unsigned long) -EINVAL;
1779
David Howells8feae132009-01-08 12:04:47 +00001780 vma = find_vma_exact(current->mm, addr, old_len);
David Howells6fa5f802006-09-27 01:50:21 -07001781 if (!vma)
1782 return (unsigned long) -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783
David Howells6fa5f802006-09-27 01:50:21 -07001784 if (vma->vm_end != vma->vm_start + old_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785 return (unsigned long) -EFAULT;
1786
David Howells6fa5f802006-09-27 01:50:21 -07001787 if (vma->vm_flags & VM_MAYSHARE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788 return (unsigned long) -EPERM;
1789
David Howells8feae132009-01-08 12:04:47 +00001790 if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791 return (unsigned long) -ENOMEM;
1792
1793 /* all checks complete - do it */
David Howells6fa5f802006-09-27 01:50:21 -07001794 vma->vm_end = vma->vm_start + new_len;
David Howells6fa5f802006-09-27 01:50:21 -07001795 return vma->vm_start;
1796}
Paul Mundtb5073172007-07-21 04:37:25 -07001797EXPORT_SYMBOL(do_mremap);
David Howells6fa5f802006-09-27 01:50:21 -07001798
Heiko Carstens6a6160a2009-01-14 14:14:15 +01001799SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1800 unsigned long, new_len, unsigned long, flags,
1801 unsigned long, new_addr)
David Howells6fa5f802006-09-27 01:50:21 -07001802{
1803 unsigned long ret;
1804
1805 down_write(&current->mm->mmap_sem);
1806 ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1807 up_write(&current->mm->mmap_sem);
1808 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809}
1810
Linus Torvalds6aab3412005-11-28 14:34:23 -08001811struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001812 unsigned int foll_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813{
1814 return NULL;
1815}
1816
Bob Liu8f3b1322011-07-08 15:39:46 -07001817int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1818 unsigned long pfn, unsigned long size, pgprot_t prot)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819{
Bob Liu8f3b1322011-07-08 15:39:46 -07001820 if (addr != (pfn << PAGE_SHIFT))
1821 return -EINVAL;
1822
1823 vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP;
Greg Ungerer66aa2b42005-09-12 11:18:10 +10001824 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825}
Luke Yang22c4af42006-07-14 00:24:09 -07001826EXPORT_SYMBOL(remap_pfn_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827
Paul Mundtf905bc42008-02-04 22:29:59 -08001828int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1829 unsigned long pgoff)
1830{
1831 unsigned int size = vma->vm_end - vma->vm_start;
1832
1833 if (!(vma->vm_flags & VM_USERMAP))
1834 return -EINVAL;
1835
1836 vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1837 vma->vm_end = vma->vm_start + size;
1838
1839 return 0;
1840}
1841EXPORT_SYMBOL(remap_vmalloc_range);
1842
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1844 unsigned long len, unsigned long pgoff, unsigned long flags)
1845{
1846 return -ENOMEM;
1847}
1848
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001849void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850{
1851}
1852
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853void unmap_mapping_range(struct address_space *mapping,
1854 loff_t const holebegin, loff_t const holelen,
1855 int even_cows)
1856{
1857}
Luke Yang22c4af42006-07-14 00:24:09 -07001858EXPORT_SYMBOL(unmap_mapping_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001859
1860/*
1861 * Check that a process has enough memory to allocate a new virtual
1862 * mapping. 0 means there is enough memory for the allocation to
1863 * succeed and -ENOMEM implies there is not.
1864 *
1865 * We currently support three overcommit policies, which are set via the
1866 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
1867 *
1868 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1869 * Additional code 2002 Jul 20 by Robert Love.
1870 *
1871 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1872 *
1873 * Note this is a helper function intended to be used by LSMs which
1874 * wish to use this logic.
1875 */
Alan Cox34b4e4a2007-08-22 14:01:28 -07001876int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877{
1878 unsigned long free, allowed;
1879
1880 vm_acct_memory(pages);
1881
1882 /*
1883 * Sometimes we want to use more memory than we have
1884 */
1885 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1886 return 0;
1887
1888 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
Dmitry Finkc15bef32011-07-25 17:12:19 -07001889 free = global_page_state(NR_FREE_PAGES);
1890 free += global_page_state(NR_FILE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891
Dmitry Finkc15bef32011-07-25 17:12:19 -07001892 /*
1893 * shmem pages shouldn't be counted as free in this
1894 * case, they can't be purged, only swapped out, and
1895 * that won't affect the overall amount of available
1896 * memory in the system.
1897 */
1898 free -= global_page_state(NR_SHMEM);
1899
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900 free += nr_swap_pages;
1901
1902 /*
1903 * Any slabs which are created with the
1904 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1905 * which are reclaimable, under pressure. The dentry
1906 * cache and most inode caches should fall into this
1907 */
Christoph Lameter972d1a72006-09-25 23:31:51 -07001908 free += global_page_state(NR_SLAB_RECLAIMABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909
1910 /*
Dmitry Finkc15bef32011-07-25 17:12:19 -07001911 * Leave reserved pages. The pages are not for anonymous pages.
1912 */
1913 if (free <= totalreserve_pages)
1914 goto error;
1915 else
1916 free -= totalreserve_pages;
1917
1918 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919 * Leave the last 3% for root
1920 */
1921 if (!cap_sys_admin)
1922 free -= free / 32;
1923
1924 if (free > pages)
1925 return 0;
1926
Hideo AOKId5ddc792006-04-10 22:53:01 -07001927 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928 }
1929
1930 allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1931 /*
1932 * Leave the last 3% for root
1933 */
1934 if (!cap_sys_admin)
1935 allowed -= allowed / 32;
1936 allowed += total_swap_pages;
1937
1938 /* Don't let a single process grow too big:
1939 leave 3% of the size of this process for other processes */
Alan Cox731572d2008-10-29 14:01:20 -07001940 if (mm)
1941 allowed -= mm->total_vm / 32;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07001943 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944 return 0;
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07001945
Hideo AOKId5ddc792006-04-10 22:53:01 -07001946error:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947 vm_unacct_memory(pages);
1948
1949 return -ENOMEM;
1950}
1951
Stephen Wilsoncae5d392011-03-13 15:49:17 -04001952int in_gate_area_no_mm(unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953{
1954 return 0;
1955}
David Howellsb0e15192006-01-06 00:11:42 -08001956
Nick Piggind0217ac2007-07-19 01:47:03 -07001957int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
David Howellsb0e15192006-01-06 00:11:42 -08001958{
1959 BUG();
Nick Piggind0217ac2007-07-19 01:47:03 -07001960 return 0;
David Howellsb0e15192006-01-06 00:11:42 -08001961}
Paul Mundtb5073172007-07-21 04:37:25 -07001962EXPORT_SYMBOL(filemap_fault);
David Howells0ec76a12006-09-27 01:50:15 -07001963
Mike Frysingerf55f1992011-03-29 14:05:12 +01001964static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
1965 unsigned long addr, void *buf, int len, int write)
David Howells0ec76a12006-09-27 01:50:15 -07001966{
David Howells0ec76a12006-09-27 01:50:15 -07001967 struct vm_area_struct *vma;
David Howells0ec76a12006-09-27 01:50:15 -07001968
1969 down_read(&mm->mmap_sem);
1970
1971 /* the access must start within one of the target process's mappings */
David Howells0159b142006-09-27 01:50:16 -07001972 vma = find_vma(mm, addr);
1973 if (vma) {
David Howells0ec76a12006-09-27 01:50:15 -07001974 /* don't overrun this mapping */
1975 if (addr + len >= vma->vm_end)
1976 len = vma->vm_end - addr;
1977
1978 /* only read or write mappings where it is permitted */
David Howellsd00c7b992006-09-27 01:50:19 -07001979 if (write && vma->vm_flags & VM_MAYWRITE)
Jie Zhang79597222010-01-06 17:23:28 +00001980 copy_to_user_page(vma, NULL, addr,
1981 (void *) addr, buf, len);
David Howellsd00c7b992006-09-27 01:50:19 -07001982 else if (!write && vma->vm_flags & VM_MAYREAD)
Jie Zhang79597222010-01-06 17:23:28 +00001983 copy_from_user_page(vma, NULL, addr,
1984 buf, (void *) addr, len);
David Howells0ec76a12006-09-27 01:50:15 -07001985 else
1986 len = 0;
1987 } else {
1988 len = 0;
1989 }
1990
1991 up_read(&mm->mmap_sem);
Mike Frysingerf55f1992011-03-29 14:05:12 +01001992
1993 return len;
1994}
1995
1996/**
1997 * @access_remote_vm - access another process' address space
1998 * @mm: the mm_struct of the target address space
1999 * @addr: start address to access
2000 * @buf: source or destination buffer
2001 * @len: number of bytes to transfer
2002 * @write: whether the access is a write
2003 *
2004 * The caller must hold a reference on @mm.
2005 */
2006int access_remote_vm(struct mm_struct *mm, unsigned long addr,
2007 void *buf, int len, int write)
2008{
2009 return __access_remote_vm(NULL, mm, addr, buf, len, write);
2010}
2011
2012/*
2013 * Access another process' address space.
2014 * - source/target buffer must be kernel space
2015 */
2016int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
2017{
2018 struct mm_struct *mm;
2019
2020 if (addr + len < addr)
2021 return 0;
2022
2023 mm = get_task_mm(tsk);
2024 if (!mm)
2025 return 0;
2026
2027 len = __access_remote_vm(tsk, mm, addr, buf, len, write);
2028
David Howells0ec76a12006-09-27 01:50:15 -07002029 mmput(mm);
2030 return len;
2031}
David Howells7e660872010-01-15 17:01:39 -08002032
2033/**
2034 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
2035 * @inode: The inode to check
2036 * @size: The current filesize of the inode
2037 * @newsize: The proposed filesize of the inode
2038 *
2039 * Check the shared mappings on an inode on behalf of a shrinking truncate to
2040 * make sure that that any outstanding VMAs aren't broken and then shrink the
2041 * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
2042 * automatically grant mappings that are too large.
2043 */
2044int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
2045 size_t newsize)
2046{
2047 struct vm_area_struct *vma;
2048 struct prio_tree_iter iter;
2049 struct vm_region *region;
2050 pgoff_t low, high;
2051 size_t r_size, r_top;
2052
2053 low = newsize >> PAGE_SHIFT;
2054 high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
2055
2056 down_write(&nommu_region_sem);
David Howells918e5562012-02-23 13:50:35 +00002057 mutex_lock(&inode->i_mapping->i_mmap_mutex);
David Howells7e660872010-01-15 17:01:39 -08002058
2059 /* search for VMAs that fall within the dead zone */
2060 vma_prio_tree_foreach(vma, &iter, &inode->i_mapping->i_mmap,
2061 low, high) {
2062 /* found one - only interested if it's shared out of the page
2063 * cache */
2064 if (vma->vm_flags & VM_SHARED) {
David Howells918e5562012-02-23 13:50:35 +00002065 mutex_unlock(&inode->i_mapping->i_mmap_mutex);
David Howells7e660872010-01-15 17:01:39 -08002066 up_write(&nommu_region_sem);
2067 return -ETXTBSY; /* not quite true, but near enough */
2068 }
2069 }
2070
2071 /* reduce any regions that overlap the dead zone - if in existence,
2072 * these will be pointed to by VMAs that don't overlap the dead zone
2073 *
2074 * we don't check for any regions that start beyond the EOF as there
2075 * shouldn't be any
2076 */
2077 vma_prio_tree_foreach(vma, &iter, &inode->i_mapping->i_mmap,
2078 0, ULONG_MAX) {
2079 if (!(vma->vm_flags & VM_SHARED))
2080 continue;
2081
2082 region = vma->vm_region;
2083 r_size = region->vm_top - region->vm_start;
2084 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
2085
2086 if (r_top > newsize) {
2087 region->vm_top -= r_top - newsize;
2088 if (region->vm_end > region->vm_top)
2089 region->vm_end = region->vm_top;
2090 }
2091 }
2092
David Howells918e5562012-02-23 13:50:35 +00002093 mutex_unlock(&inode->i_mapping->i_mmap_mutex);
David Howells7e660872010-01-15 17:01:39 -08002094 up_write(&nommu_region_sem);
2095 return 0;
2096}