blob: c4c542c736a962774f0770eef0d50c419b19a2cb [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
David Howellsf2b85442007-10-29 13:15:39 +000016#include <linux/module.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>
Roland McGrathfa8e26c2008-07-25 19:45:50 -070025#include <linux/tracehook.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/blkdev.h>
27#include <linux/backing-dev.h>
28#include <linux/mount.h>
29#include <linux/personality.h>
30#include <linux/security.h>
31#include <linux/syscalls.h>
Al Viro120a7952010-10-30 02:54:44 -040032#include <linux/audit.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070033
34#include <asm/uaccess.h>
35#include <asm/tlb.h>
36#include <asm/tlbflush.h>
Bernd Schmidteb8cdec2009-09-21 17:03:57 -070037#include <asm/mmu_context.h>
David Howells8feae132009-01-08 12:04:47 +000038#include "internal.h"
39
David Howells8feae132009-01-08 12:04:47 +000040#if 0
41#define kenter(FMT, ...) \
42 printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
43#define kleave(FMT, ...) \
44 printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
45#define kdebug(FMT, ...) \
46 printk(KERN_DEBUG "xxx" FMT"yyy\n", ##__VA_ARGS__)
47#else
48#define kenter(FMT, ...) \
49 no_printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
50#define kleave(FMT, ...) \
51 no_printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
52#define kdebug(FMT, ...) \
53 no_printk(KERN_DEBUG FMT"\n", ##__VA_ARGS__)
54#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -070055
56void *high_memory;
57struct page *mem_map;
58unsigned long max_mapnr;
59unsigned long num_physpages;
Hugh Dickins4266c972009-09-23 17:05:53 +010060unsigned long highest_memmap_pfn;
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -070061struct percpu_counter vm_committed_as;
Linus Torvalds1da177e2005-04-16 15:20:36 -070062int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
63int sysctl_overcommit_ratio = 50; /* default is 50% */
64int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
David Howellsfc4d5c22009-05-06 16:03:05 -070065int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -070066int heap_stack_gap = 0;
67
David Howells33e5d7692009-04-02 16:56:32 -070068atomic_long_t mmap_pages_allocated;
David Howells8feae132009-01-08 12:04:47 +000069
Linus Torvalds1da177e2005-04-16 15:20:36 -070070EXPORT_SYMBOL(mem_map);
Wu, Bryan6a04de62007-04-11 23:28:47 -070071EXPORT_SYMBOL(num_physpages);
Linus Torvalds1da177e2005-04-16 15:20:36 -070072
David Howells8feae132009-01-08 12:04:47 +000073/* list of mapped, potentially shareable regions */
74static struct kmem_cache *vm_region_jar;
75struct rb_root nommu_region_tree = RB_ROOT;
76DECLARE_RWSEM(nommu_region_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -070077
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +040078const struct vm_operations_struct generic_file_vm_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -070079};
80
81/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070082 * Return the total memory allocated for this pointer, not
83 * just what the caller asked for.
84 *
85 * Doesn't have to be accurate, i.e. may have races.
86 */
87unsigned int kobjsize(const void *objp)
88{
89 struct page *page;
90
Michael Hennerich4016a132008-04-28 02:13:38 -070091 /*
92 * If the object we have should not have ksize performed on it,
93 * return size of 0
94 */
Paul Mundt5a1603b2008-06-12 16:29:55 +090095 if (!objp || !virt_addr_valid(objp))
Paul Mundt6cfd53fc2008-06-05 22:46:08 -070096 return 0;
97
98 page = virt_to_head_page(objp);
Paul Mundt6cfd53fc2008-06-05 22:46:08 -070099
100 /*
101 * If the allocator sets PageSlab, we know the pointer came from
102 * kmalloc().
103 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104 if (PageSlab(page))
105 return ksize(objp);
106
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700107 /*
Paul Mundtab2e83e2009-01-08 12:04:48 +0000108 * If it's not a compound page, see if we have a matching VMA
109 * region. This test is intentionally done in reverse order,
110 * so if there's no VMA, we still fall through and hand back
111 * PAGE_SIZE for 0-order pages.
112 */
113 if (!PageCompound(page)) {
114 struct vm_area_struct *vma;
115
116 vma = find_vma(current->mm, (unsigned long)objp);
117 if (vma)
118 return vma->vm_end - vma->vm_start;
119 }
120
121 /*
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700122 * The ksize() function is only guaranteed to work for pointers
Paul Mundt5a1603b2008-06-12 16:29:55 +0900123 * returned by kmalloc(). So handle arbitrary pointers here.
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700124 */
Paul Mundt5a1603b2008-06-12 16:29:55 +0900125 return PAGE_SIZE << compound_order(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126}
127
Nick Pigginb291f002008-10-18 20:26:44 -0700128int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
Hugh Dickins4266c972009-09-23 17:05:53 +0100129 unsigned long start, int nr_pages, unsigned int foll_flags,
Michel Lespinasse53a77062011-01-13 15:46:14 -0800130 struct page **pages, struct vm_area_struct **vmas,
131 int *retry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132{
Sonic Zhang910e46d2006-09-27 01:50:17 -0700133 struct vm_area_struct *vma;
David Howells7b4d5b82006-09-27 01:50:18 -0700134 unsigned long vm_flags;
135 int i;
136
137 /* calculate required read or write permissions.
Hugh Dickins58fa8792009-09-21 17:03:31 -0700138 * If FOLL_FORCE is set, we only require the "MAY" flags.
David Howells7b4d5b82006-09-27 01:50:18 -0700139 */
Hugh Dickins58fa8792009-09-21 17:03:31 -0700140 vm_flags = (foll_flags & FOLL_WRITE) ?
141 (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
142 vm_flags &= (foll_flags & FOLL_FORCE) ?
143 (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144
Peter Zijlstra9d737772009-06-25 11:58:55 +0200145 for (i = 0; i < nr_pages; i++) {
David Howells7561e8c2010-03-25 16:48:38 +0000146 vma = find_vma(mm, start);
David Howells7b4d5b82006-09-27 01:50:18 -0700147 if (!vma)
148 goto finish_or_fault;
149
150 /* protect what we can, including chardevs */
Hugh Dickins1c3aff12009-09-21 17:03:24 -0700151 if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
152 !(vm_flags & vma->vm_flags))
David Howells7b4d5b82006-09-27 01:50:18 -0700153 goto finish_or_fault;
Sonic Zhang910e46d2006-09-27 01:50:17 -0700154
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 if (pages) {
156 pages[i] = virt_to_page(start);
157 if (pages[i])
158 page_cache_get(pages[i]);
159 }
160 if (vmas)
Sonic Zhang910e46d2006-09-27 01:50:17 -0700161 vmas[i] = vma;
David Howellse1ee65d2010-03-25 16:48:44 +0000162 start = (start + PAGE_SIZE) & PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 }
David Howells7b4d5b82006-09-27 01:50:18 -0700164
165 return i;
166
167finish_or_fault:
168 return i ? : -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169}
Nick Pigginb291f002008-10-18 20:26:44 -0700170
Nick Pigginb291f002008-10-18 20:26:44 -0700171/*
172 * get a list of pages in an address range belonging to the specified process
173 * and indicate the VMA that covers each page
174 * - this is potentially dodgy as we may end incrementing the page count of a
175 * slab page or a secondary page from a compound page
176 * - don't permit access to VMAs that don't support it, such as I/O mappings
177 */
178int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
Peter Zijlstra9d737772009-06-25 11:58:55 +0200179 unsigned long start, int nr_pages, int write, int force,
Nick Pigginb291f002008-10-18 20:26:44 -0700180 struct page **pages, struct vm_area_struct **vmas)
181{
182 int flags = 0;
183
184 if (write)
Hugh Dickins58fa8792009-09-21 17:03:31 -0700185 flags |= FOLL_WRITE;
Nick Pigginb291f002008-10-18 20:26:44 -0700186 if (force)
Hugh Dickins58fa8792009-09-21 17:03:31 -0700187 flags |= FOLL_FORCE;
Nick Pigginb291f002008-10-18 20:26:44 -0700188
Michel Lespinasse53a77062011-01-13 15:46:14 -0800189 return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas,
190 NULL);
Nick Pigginb291f002008-10-18 20:26:44 -0700191}
Greg Ungerer66aa2b42005-09-12 11:18:10 +1000192EXPORT_SYMBOL(get_user_pages);
193
Paul Mundtdfc2f912009-06-26 04:31:57 +0900194/**
195 * follow_pfn - look up PFN at a user virtual address
196 * @vma: memory mapping
197 * @address: user virtual address
198 * @pfn: location to store found PFN
199 *
200 * Only IO mappings and raw PFN mappings are allowed.
201 *
202 * Returns zero and the pfn at @pfn on success, -ve otherwise.
203 */
204int follow_pfn(struct vm_area_struct *vma, unsigned long address,
205 unsigned long *pfn)
206{
207 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
208 return -EINVAL;
209
210 *pfn = address >> PAGE_SHIFT;
211 return 0;
212}
213EXPORT_SYMBOL(follow_pfn);
214
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215DEFINE_RWLOCK(vmlist_lock);
216struct vm_struct *vmlist;
217
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800218void vfree(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219{
220 kfree(addr);
221}
Paul Mundtb5073172007-07-21 04:37:25 -0700222EXPORT_SYMBOL(vfree);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223
Al Virodd0fc662005-10-07 07:46:04 +0100224void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225{
226 /*
Robert P. J. Day85186092007-10-19 23:11:38 +0200227 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
228 * returns only a logical address.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 */
Nick Piggin84097512006-03-22 00:08:34 -0800230 return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231}
Paul Mundtb5073172007-07-21 04:37:25 -0700232EXPORT_SYMBOL(__vmalloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233
Paul Mundtf905bc42008-02-04 22:29:59 -0800234void *vmalloc_user(unsigned long size)
235{
236 void *ret;
237
238 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
239 PAGE_KERNEL);
240 if (ret) {
241 struct vm_area_struct *vma;
242
243 down_write(&current->mm->mmap_sem);
244 vma = find_vma(current->mm, (unsigned long)ret);
245 if (vma)
246 vma->vm_flags |= VM_USERMAP;
247 up_write(&current->mm->mmap_sem);
248 }
249
250 return ret;
251}
252EXPORT_SYMBOL(vmalloc_user);
253
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800254struct page *vmalloc_to_page(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255{
256 return virt_to_page(addr);
257}
Paul Mundtb5073172007-07-21 04:37:25 -0700258EXPORT_SYMBOL(vmalloc_to_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800260unsigned long vmalloc_to_pfn(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261{
262 return page_to_pfn(virt_to_page(addr));
263}
Paul Mundtb5073172007-07-21 04:37:25 -0700264EXPORT_SYMBOL(vmalloc_to_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265
266long vread(char *buf, char *addr, unsigned long count)
267{
268 memcpy(buf, addr, count);
269 return count;
270}
271
272long vwrite(char *buf, char *addr, unsigned long count)
273{
274 /* Don't allow overflow */
275 if ((unsigned long) addr + count < count)
276 count = -(unsigned long) addr;
277
278 memcpy(addr, buf, count);
279 return(count);
280}
281
282/*
283 * vmalloc - allocate virtually continguos memory
284 *
285 * @size: allocation size
286 *
287 * Allocate enough pages to cover @size from the page level
288 * allocator and map them into continguos kernel virtual space.
289 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +0200290 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 * use __vmalloc() instead.
292 */
293void *vmalloc(unsigned long size)
294{
295 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
296}
Andrew Mortonf6138882006-02-28 16:59:18 -0800297EXPORT_SYMBOL(vmalloc);
298
Dave Younge1ca7782010-10-26 14:22:06 -0700299/*
300 * vzalloc - allocate virtually continguos memory with zero fill
301 *
302 * @size: allocation size
303 *
304 * Allocate enough pages to cover @size from the page level
305 * allocator and map them into continguos kernel virtual space.
306 * The memory allocated is set to zero.
307 *
308 * For tight control over page level allocator and protection flags
309 * use __vmalloc() instead.
310 */
311void *vzalloc(unsigned long size)
312{
313 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
314 PAGE_KERNEL);
315}
316EXPORT_SYMBOL(vzalloc);
317
318/**
319 * vmalloc_node - allocate memory on a specific node
320 * @size: allocation size
321 * @node: numa node
322 *
323 * Allocate enough pages to cover @size from the page level
324 * allocator and map them into contiguous kernel virtual space.
325 *
326 * For tight control over page level allocator and protection flags
327 * use __vmalloc() instead.
328 */
Andrew Mortonf6138882006-02-28 16:59:18 -0800329void *vmalloc_node(unsigned long size, int node)
330{
331 return vmalloc(size);
332}
Paul Mundt9a14f652010-12-24 11:50:34 +0900333EXPORT_SYMBOL(vmalloc_node);
Dave Younge1ca7782010-10-26 14:22:06 -0700334
335/**
336 * vzalloc_node - allocate memory on a specific node with zero fill
337 * @size: allocation size
338 * @node: numa node
339 *
340 * Allocate enough pages to cover @size from the page level
341 * allocator and map them into contiguous kernel virtual space.
342 * The memory allocated is set to zero.
343 *
344 * For tight control over page level allocator and protection flags
345 * use __vmalloc() instead.
346 */
347void *vzalloc_node(unsigned long size, int node)
348{
349 return vzalloc(size);
350}
351EXPORT_SYMBOL(vzalloc_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352
Paul Mundt1af446e2008-08-04 16:01:47 +0900353#ifndef PAGE_KERNEL_EXEC
354# define PAGE_KERNEL_EXEC PAGE_KERNEL
355#endif
356
357/**
358 * vmalloc_exec - allocate virtually contiguous, executable memory
359 * @size: allocation size
360 *
361 * Kernel-internal function to allocate enough pages to cover @size
362 * the page level allocator and map them into contiguous and
363 * executable kernel virtual space.
364 *
365 * For tight control over page level allocator and protection flags
366 * use __vmalloc() instead.
367 */
368
369void *vmalloc_exec(unsigned long size)
370{
371 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
372}
373
Paul Mundtb5073172007-07-21 04:37:25 -0700374/**
375 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 * @size: allocation size
377 *
378 * Allocate enough 32bit PA addressable pages to cover @size from the
379 * page level allocator and map them into continguos kernel virtual space.
380 */
381void *vmalloc_32(unsigned long size)
382{
383 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
384}
Paul Mundtb5073172007-07-21 04:37:25 -0700385EXPORT_SYMBOL(vmalloc_32);
386
387/**
388 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
389 * @size: allocation size
390 *
391 * The resulting memory area is 32bit addressable and zeroed so it can be
392 * mapped to userspace without leaking data.
Paul Mundtf905bc42008-02-04 22:29:59 -0800393 *
394 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
395 * remap_vmalloc_range() are permissible.
Paul Mundtb5073172007-07-21 04:37:25 -0700396 */
397void *vmalloc_32_user(unsigned long size)
398{
Paul Mundtf905bc42008-02-04 22:29:59 -0800399 /*
400 * We'll have to sort out the ZONE_DMA bits for 64-bit,
401 * but for now this can simply use vmalloc_user() directly.
402 */
403 return vmalloc_user(size);
Paul Mundtb5073172007-07-21 04:37:25 -0700404}
405EXPORT_SYMBOL(vmalloc_32_user);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406
407void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
408{
409 BUG();
410 return NULL;
411}
Paul Mundtb5073172007-07-21 04:37:25 -0700412EXPORT_SYMBOL(vmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800414void vunmap(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415{
416 BUG();
417}
Paul Mundtb5073172007-07-21 04:37:25 -0700418EXPORT_SYMBOL(vunmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419
Paul Mundteb6434d2009-01-21 17:45:47 +0900420void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
421{
422 BUG();
423 return NULL;
424}
425EXPORT_SYMBOL(vm_map_ram);
426
427void vm_unmap_ram(const void *mem, unsigned int count)
428{
429 BUG();
430}
431EXPORT_SYMBOL(vm_unmap_ram);
432
433void vm_unmap_aliases(void)
434{
435}
436EXPORT_SYMBOL_GPL(vm_unmap_aliases);
437
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438/*
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -0700439 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
440 * have one.
441 */
442void __attribute__((weak)) vmalloc_sync_all(void)
443{
444}
445
Paul Mundt29c185e2010-12-24 12:08:30 +0900446/**
447 * alloc_vm_area - allocate a range of kernel address space
448 * @size: size of the area
449 *
450 * Returns: NULL on failure, vm_struct on success
451 *
452 * This function reserves a range of kernel address space, and
453 * allocates pagetables to map that range. No actual mappings
454 * are created. If the kernel address space is not shared
455 * between processes, it syncs the pagetable across all
456 * processes.
457 */
458struct vm_struct *alloc_vm_area(size_t size)
459{
460 BUG();
461 return NULL;
462}
463EXPORT_SYMBOL_GPL(alloc_vm_area);
464
465void free_vm_area(struct vm_struct *area)
466{
467 BUG();
468}
469EXPORT_SYMBOL_GPL(free_vm_area);
470
Paul Mundtb5073172007-07-21 04:37:25 -0700471int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
472 struct page *page)
473{
474 return -EINVAL;
475}
476EXPORT_SYMBOL(vm_insert_page);
477
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -0700478/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 * sys_brk() for the most part doesn't need the global kernel
480 * lock, except when an application is doing something nasty
481 * like trying to un-brk an area that has already been mapped
482 * to a regular file. in this case, the unmapping will need
483 * to invoke file system routines that need the global lock.
484 */
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100485SYSCALL_DEFINE1(brk, unsigned long, brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486{
487 struct mm_struct *mm = current->mm;
488
489 if (brk < mm->start_brk || brk > mm->context.end_brk)
490 return mm->brk;
491
492 if (mm->brk == brk)
493 return mm->brk;
494
495 /*
496 * Always allow shrinking brk
497 */
498 if (brk <= mm->brk) {
499 mm->brk = brk;
500 return brk;
501 }
502
503 /*
504 * Ok, looks good - let it rip.
505 */
Mike Frysingercfe79c02010-01-06 17:23:23 +0000506 flush_icache_range(mm->brk, brk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 return mm->brk = brk;
508}
509
David Howells8feae132009-01-08 12:04:47 +0000510/*
511 * initialise the VMA and region record slabs
512 */
513void __init mmap_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514{
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -0700515 int ret;
516
517 ret = percpu_counter_init(&vm_committed_as, 0);
518 VM_BUG_ON(ret);
David Howells33e5d7692009-04-02 16:56:32 -0700519 vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC);
David Howells8feae132009-01-08 12:04:47 +0000520}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521
David Howells8feae132009-01-08 12:04:47 +0000522/*
523 * validate the region tree
524 * - the caller must hold the region lock
525 */
526#ifdef CONFIG_DEBUG_NOMMU_REGIONS
527static noinline void validate_nommu_regions(void)
528{
529 struct vm_region *region, *last;
530 struct rb_node *p, *lastp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531
David Howells8feae132009-01-08 12:04:47 +0000532 lastp = rb_first(&nommu_region_tree);
533 if (!lastp)
534 return;
535
536 last = rb_entry(lastp, struct vm_region, vm_rb);
David Howells33e5d7692009-04-02 16:56:32 -0700537 BUG_ON(unlikely(last->vm_end <= last->vm_start));
538 BUG_ON(unlikely(last->vm_top < last->vm_end));
David Howells8feae132009-01-08 12:04:47 +0000539
540 while ((p = rb_next(lastp))) {
541 region = rb_entry(p, struct vm_region, vm_rb);
542 last = rb_entry(lastp, struct vm_region, vm_rb);
543
David Howells33e5d7692009-04-02 16:56:32 -0700544 BUG_ON(unlikely(region->vm_end <= region->vm_start));
545 BUG_ON(unlikely(region->vm_top < region->vm_end));
546 BUG_ON(unlikely(region->vm_start < last->vm_top));
David Howells8feae132009-01-08 12:04:47 +0000547
548 lastp = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 }
550}
David Howells8feae132009-01-08 12:04:47 +0000551#else
David Howells33e5d7692009-04-02 16:56:32 -0700552static void validate_nommu_regions(void)
553{
554}
David Howells8feae132009-01-08 12:04:47 +0000555#endif
556
557/*
558 * add a region into the global tree
559 */
560static void add_nommu_region(struct vm_region *region)
561{
562 struct vm_region *pregion;
563 struct rb_node **p, *parent;
564
565 validate_nommu_regions();
566
David Howells8feae132009-01-08 12:04:47 +0000567 parent = NULL;
568 p = &nommu_region_tree.rb_node;
569 while (*p) {
570 parent = *p;
571 pregion = rb_entry(parent, struct vm_region, vm_rb);
572 if (region->vm_start < pregion->vm_start)
573 p = &(*p)->rb_left;
574 else if (region->vm_start > pregion->vm_start)
575 p = &(*p)->rb_right;
576 else if (pregion == region)
577 return;
578 else
579 BUG();
580 }
581
582 rb_link_node(&region->vm_rb, parent, p);
583 rb_insert_color(&region->vm_rb, &nommu_region_tree);
584
585 validate_nommu_regions();
586}
587
588/*
589 * delete a region from the global tree
590 */
591static void delete_nommu_region(struct vm_region *region)
592{
593 BUG_ON(!nommu_region_tree.rb_node);
594
595 validate_nommu_regions();
596 rb_erase(&region->vm_rb, &nommu_region_tree);
597 validate_nommu_regions();
598}
599
600/*
601 * free a contiguous series of pages
602 */
603static void free_page_series(unsigned long from, unsigned long to)
604{
605 for (; from < to; from += PAGE_SIZE) {
606 struct page *page = virt_to_page(from);
607
608 kdebug("- free %lx", from);
David Howells33e5d7692009-04-02 16:56:32 -0700609 atomic_long_dec(&mmap_pages_allocated);
David Howells8feae132009-01-08 12:04:47 +0000610 if (page_count(page) != 1)
David Howells33e5d7692009-04-02 16:56:32 -0700611 kdebug("free page %p: refcount not one: %d",
612 page, page_count(page));
David Howells8feae132009-01-08 12:04:47 +0000613 put_page(page);
614 }
615}
616
617/*
618 * release a reference to a region
David Howells33e5d7692009-04-02 16:56:32 -0700619 * - the caller must hold the region semaphore for writing, which this releases
Paul Mundtdd8632a2009-01-08 12:04:47 +0000620 * - the region may not have been added to the tree yet, in which case vm_top
David Howells8feae132009-01-08 12:04:47 +0000621 * will equal vm_start
622 */
623static void __put_nommu_region(struct vm_region *region)
624 __releases(nommu_region_sem)
625{
David Howells1e2ae592010-01-15 17:01:33 -0800626 kenter("%p{%d}", region, region->vm_usage);
David Howells8feae132009-01-08 12:04:47 +0000627
628 BUG_ON(!nommu_region_tree.rb_node);
629
David Howells1e2ae592010-01-15 17:01:33 -0800630 if (--region->vm_usage == 0) {
Paul Mundtdd8632a2009-01-08 12:04:47 +0000631 if (region->vm_top > region->vm_start)
David Howells8feae132009-01-08 12:04:47 +0000632 delete_nommu_region(region);
633 up_write(&nommu_region_sem);
634
635 if (region->vm_file)
636 fput(region->vm_file);
637
638 /* IO memory and memory shared directly out of the pagecache
639 * from ramfs/tmpfs mustn't be released here */
640 if (region->vm_flags & VM_MAPPED_COPY) {
641 kdebug("free series");
Paul Mundtdd8632a2009-01-08 12:04:47 +0000642 free_page_series(region->vm_start, region->vm_top);
David Howells8feae132009-01-08 12:04:47 +0000643 }
644 kmem_cache_free(vm_region_jar, region);
645 } else {
646 up_write(&nommu_region_sem);
647 }
648}
649
650/*
651 * release a reference to a region
652 */
653static void put_nommu_region(struct vm_region *region)
654{
655 down_write(&nommu_region_sem);
656 __put_nommu_region(region);
657}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658
David Howells30340972006-09-27 01:50:20 -0700659/*
Bernd Schmidteb8cdec2009-09-21 17:03:57 -0700660 * update protection on a vma
661 */
662static void protect_vma(struct vm_area_struct *vma, unsigned long flags)
663{
664#ifdef CONFIG_MPU
665 struct mm_struct *mm = vma->vm_mm;
666 long start = vma->vm_start & PAGE_MASK;
667 while (start < vma->vm_end) {
668 protect_page(mm, start, flags);
669 start += PAGE_SIZE;
670 }
671 update_protections(mm);
672#endif
673}
674
675/*
David Howells30340972006-09-27 01:50:20 -0700676 * add a VMA into a process's mm_struct in the appropriate place in the list
David Howells8feae132009-01-08 12:04:47 +0000677 * and tree and add to the address space's page tree also if not an anonymous
678 * page
David Howells30340972006-09-27 01:50:20 -0700679 * - should be called with mm->mmap_sem held writelocked
680 */
David Howells8feae132009-01-08 12:04:47 +0000681static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
David Howells30340972006-09-27 01:50:20 -0700682{
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700683 struct vm_area_struct *pvma, **pp, *next;
David Howells8feae132009-01-08 12:04:47 +0000684 struct address_space *mapping;
685 struct rb_node **p, *parent;
David Howells30340972006-09-27 01:50:20 -0700686
David Howells8feae132009-01-08 12:04:47 +0000687 kenter(",%p", vma);
688
689 BUG_ON(!vma->vm_region);
690
691 mm->map_count++;
692 vma->vm_mm = mm;
693
Bernd Schmidteb8cdec2009-09-21 17:03:57 -0700694 protect_vma(vma, vma->vm_flags);
695
David Howells8feae132009-01-08 12:04:47 +0000696 /* add the VMA to the mapping */
697 if (vma->vm_file) {
698 mapping = vma->vm_file->f_mapping;
699
700 flush_dcache_mmap_lock(mapping);
701 vma_prio_tree_insert(vma, &mapping->i_mmap);
702 flush_dcache_mmap_unlock(mapping);
703 }
704
705 /* add the VMA to the tree */
706 parent = NULL;
707 p = &mm->mm_rb.rb_node;
708 while (*p) {
709 parent = *p;
710 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
711
712 /* sort by: start addr, end addr, VMA struct addr in that order
713 * (the latter is necessary as we may get identical VMAs) */
714 if (vma->vm_start < pvma->vm_start)
715 p = &(*p)->rb_left;
716 else if (vma->vm_start > pvma->vm_start)
717 p = &(*p)->rb_right;
718 else if (vma->vm_end < pvma->vm_end)
719 p = &(*p)->rb_left;
720 else if (vma->vm_end > pvma->vm_end)
721 p = &(*p)->rb_right;
722 else if (vma < pvma)
723 p = &(*p)->rb_left;
724 else if (vma > pvma)
725 p = &(*p)->rb_right;
726 else
727 BUG();
728 }
729
730 rb_link_node(&vma->vm_rb, parent, p);
731 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
732
733 /* add VMA to the VMA list also */
734 for (pp = &mm->mmap; (pvma = *pp); pp = &(*pp)->vm_next) {
735 if (pvma->vm_start > vma->vm_start)
David Howells30340972006-09-27 01:50:20 -0700736 break;
David Howells8feae132009-01-08 12:04:47 +0000737 if (pvma->vm_start < vma->vm_start)
738 continue;
739 if (pvma->vm_end < vma->vm_end)
740 break;
741 }
David Howells30340972006-09-27 01:50:20 -0700742
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700743 next = *pp;
David Howells8feae132009-01-08 12:04:47 +0000744 *pp = vma;
Linus Torvalds297c5ee2010-08-20 16:24:55 -0700745 vma->vm_next = next;
746 if (next)
747 next->vm_prev = vma;
David Howells8feae132009-01-08 12:04:47 +0000748}
749
750/*
751 * delete a VMA from its owning mm_struct and address space
752 */
753static void delete_vma_from_mm(struct vm_area_struct *vma)
754{
755 struct vm_area_struct **pp;
756 struct address_space *mapping;
757 struct mm_struct *mm = vma->vm_mm;
758
759 kenter("%p", vma);
760
Bernd Schmidteb8cdec2009-09-21 17:03:57 -0700761 protect_vma(vma, 0);
762
David Howells8feae132009-01-08 12:04:47 +0000763 mm->map_count--;
764 if (mm->mmap_cache == vma)
765 mm->mmap_cache = NULL;
766
767 /* remove the VMA from the mapping */
768 if (vma->vm_file) {
769 mapping = vma->vm_file->f_mapping;
770
771 flush_dcache_mmap_lock(mapping);
772 vma_prio_tree_remove(vma, &mapping->i_mmap);
773 flush_dcache_mmap_unlock(mapping);
774 }
775
776 /* remove from the MM's tree and list */
777 rb_erase(&vma->vm_rb, &mm->mm_rb);
778 for (pp = &mm->mmap; *pp; pp = &(*pp)->vm_next) {
779 if (*pp == vma) {
780 *pp = vma->vm_next;
781 break;
782 }
783 }
784
785 vma->vm_mm = NULL;
786}
787
788/*
789 * destroy a VMA record
790 */
791static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
792{
793 kenter("%p", vma);
794 if (vma->vm_ops && vma->vm_ops->close)
795 vma->vm_ops->close(vma);
796 if (vma->vm_file) {
797 fput(vma->vm_file);
798 if (vma->vm_flags & VM_EXECUTABLE)
799 removed_exe_file_vma(mm);
800 }
801 put_nommu_region(vma->vm_region);
802 kmem_cache_free(vm_area_cachep, vma);
David Howells30340972006-09-27 01:50:20 -0700803}
804
805/*
806 * look up the first VMA in which addr resides, NULL if none
807 * - should be called with mm->mmap_sem at least held readlocked
808 */
809struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
810{
David Howells8feae132009-01-08 12:04:47 +0000811 struct vm_area_struct *vma;
812 struct rb_node *n = mm->mm_rb.rb_node;
David Howells30340972006-09-27 01:50:20 -0700813
David Howells8feae132009-01-08 12:04:47 +0000814 /* check the cache first */
815 vma = mm->mmap_cache;
816 if (vma && vma->vm_start <= addr && vma->vm_end > addr)
817 return vma;
818
819 /* trawl the tree (there may be multiple mappings in which addr
820 * resides) */
821 for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) {
822 vma = rb_entry(n, struct vm_area_struct, vm_rb);
823 if (vma->vm_start > addr)
824 return NULL;
825 if (vma->vm_end > addr) {
826 mm->mmap_cache = vma;
827 return vma;
828 }
David Howells30340972006-09-27 01:50:20 -0700829 }
830
David Howells30340972006-09-27 01:50:20 -0700831 return NULL;
832}
833EXPORT_SYMBOL(find_vma);
834
835/*
David Howells930e6522006-09-27 01:50:22 -0700836 * find a VMA
837 * - we don't extend stack VMAs under NOMMU conditions
838 */
839struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
840{
David Howells7561e8c2010-03-25 16:48:38 +0000841 return find_vma(mm, addr);
David Howells930e6522006-09-27 01:50:22 -0700842}
843
David Howells8feae132009-01-08 12:04:47 +0000844/*
845 * expand a stack to a given address
846 * - not supported under NOMMU conditions
847 */
Greg Ungerer57c8f632007-07-15 23:38:28 -0700848int expand_stack(struct vm_area_struct *vma, unsigned long address)
849{
850 return -ENOMEM;
851}
852
David Howells930e6522006-09-27 01:50:22 -0700853/*
David Howells6fa5f802006-09-27 01:50:21 -0700854 * look up the first VMA exactly that exactly matches addr
855 * - should be called with mm->mmap_sem at least held readlocked
856 */
David Howells8feae132009-01-08 12:04:47 +0000857static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
858 unsigned long addr,
859 unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860{
861 struct vm_area_struct *vma;
David Howells8feae132009-01-08 12:04:47 +0000862 struct rb_node *n = mm->mm_rb.rb_node;
863 unsigned long end = addr + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864
David Howells8feae132009-01-08 12:04:47 +0000865 /* check the cache first */
866 vma = mm->mmap_cache;
867 if (vma && vma->vm_start == addr && vma->vm_end == end)
868 return vma;
869
870 /* trawl the tree (there may be multiple mappings in which addr
871 * resides) */
872 for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 vma = rb_entry(n, struct vm_area_struct, vm_rb);
David Howells8feae132009-01-08 12:04:47 +0000874 if (vma->vm_start < addr)
875 continue;
876 if (vma->vm_start > addr)
877 return NULL;
878 if (vma->vm_end == end) {
879 mm->mmap_cache = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 return vma;
David Howells8feae132009-01-08 12:04:47 +0000881 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 }
883
884 return NULL;
885}
886
David Howells30340972006-09-27 01:50:20 -0700887/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 * determine whether a mapping should be permitted and, if so, what sort of
889 * mapping we're capable of supporting
890 */
891static int validate_mmap_request(struct file *file,
892 unsigned long addr,
893 unsigned long len,
894 unsigned long prot,
895 unsigned long flags,
896 unsigned long pgoff,
897 unsigned long *_capabilities)
898{
David Howells8feae132009-01-08 12:04:47 +0000899 unsigned long capabilities, rlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900 unsigned long reqprot = prot;
901 int ret;
902
903 /* do the simple checks first */
David Howells06aab5a2009-09-24 12:33:48 +0100904 if (flags & MAP_FIXED) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 printk(KERN_DEBUG
906 "%d: Can't do fixed-address/overlay mmap of RAM\n",
907 current->pid);
908 return -EINVAL;
909 }
910
911 if ((flags & MAP_TYPE) != MAP_PRIVATE &&
912 (flags & MAP_TYPE) != MAP_SHARED)
913 return -EINVAL;
914
Mike Frysingerf81cff02006-12-06 12:02:59 +1000915 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 return -EINVAL;
917
Mike Frysingerf81cff02006-12-06 12:02:59 +1000918 /* Careful about overflows.. */
David Howells8feae132009-01-08 12:04:47 +0000919 rlen = PAGE_ALIGN(len);
920 if (!rlen || rlen > TASK_SIZE)
Mike Frysingerf81cff02006-12-06 12:02:59 +1000921 return -ENOMEM;
922
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 /* offset overflow? */
David Howells8feae132009-01-08 12:04:47 +0000924 if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
Mike Frysingerf81cff02006-12-06 12:02:59 +1000925 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926
927 if (file) {
928 /* validate file mapping requests */
929 struct address_space *mapping;
930
931 /* files must support mmap */
932 if (!file->f_op || !file->f_op->mmap)
933 return -ENODEV;
934
935 /* work out if what we've got could possibly be shared
936 * - we support chardevs that provide their own "memory"
937 * - we support files/blockdevs that are memory backed
938 */
939 mapping = file->f_mapping;
940 if (!mapping)
Josef Sipeke9536ae2006-12-08 02:37:21 -0800941 mapping = file->f_path.dentry->d_inode->i_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942
943 capabilities = 0;
944 if (mapping && mapping->backing_dev_info)
945 capabilities = mapping->backing_dev_info->capabilities;
946
947 if (!capabilities) {
948 /* no explicit capabilities set, so assume some
949 * defaults */
Josef Sipeke9536ae2006-12-08 02:37:21 -0800950 switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951 case S_IFREG:
952 case S_IFBLK:
953 capabilities = BDI_CAP_MAP_COPY;
954 break;
955
956 case S_IFCHR:
957 capabilities =
958 BDI_CAP_MAP_DIRECT |
959 BDI_CAP_READ_MAP |
960 BDI_CAP_WRITE_MAP;
961 break;
962
963 default:
964 return -EINVAL;
965 }
966 }
967
968 /* eliminate any capabilities that we can't support on this
969 * device */
970 if (!file->f_op->get_unmapped_area)
971 capabilities &= ~BDI_CAP_MAP_DIRECT;
972 if (!file->f_op->read)
973 capabilities &= ~BDI_CAP_MAP_COPY;
974
Graff Yang28d7a6a2009-08-18 14:11:17 -0700975 /* The file shall have been opened with read permission. */
976 if (!(file->f_mode & FMODE_READ))
977 return -EACCES;
978
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 if (flags & MAP_SHARED) {
980 /* do checks for writing, appending and locking */
981 if ((prot & PROT_WRITE) &&
982 !(file->f_mode & FMODE_WRITE))
983 return -EACCES;
984
Josef Sipeke9536ae2006-12-08 02:37:21 -0800985 if (IS_APPEND(file->f_path.dentry->d_inode) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 (file->f_mode & FMODE_WRITE))
987 return -EACCES;
988
Josef Sipeke9536ae2006-12-08 02:37:21 -0800989 if (locks_verify_locked(file->f_path.dentry->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 return -EAGAIN;
991
992 if (!(capabilities & BDI_CAP_MAP_DIRECT))
993 return -ENODEV;
994
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 /* we mustn't privatise shared mappings */
996 capabilities &= ~BDI_CAP_MAP_COPY;
997 }
998 else {
999 /* we're going to read the file into private memory we
1000 * allocate */
1001 if (!(capabilities & BDI_CAP_MAP_COPY))
1002 return -ENODEV;
1003
1004 /* we don't permit a private writable mapping to be
1005 * shared with the backing device */
1006 if (prot & PROT_WRITE)
1007 capabilities &= ~BDI_CAP_MAP_DIRECT;
1008 }
1009
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001010 if (capabilities & BDI_CAP_MAP_DIRECT) {
1011 if (((prot & PROT_READ) && !(capabilities & BDI_CAP_READ_MAP)) ||
1012 ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
1013 ((prot & PROT_EXEC) && !(capabilities & BDI_CAP_EXEC_MAP))
1014 ) {
1015 capabilities &= ~BDI_CAP_MAP_DIRECT;
1016 if (flags & MAP_SHARED) {
1017 printk(KERN_WARNING
1018 "MAP_SHARED not completely supported on !MMU\n");
1019 return -EINVAL;
1020 }
1021 }
1022 }
1023
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 /* handle executable mappings and implied executable
1025 * mappings */
Josef Sipeke9536ae2006-12-08 02:37:21 -08001026 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 if (prot & PROT_EXEC)
1028 return -EPERM;
1029 }
1030 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
1031 /* handle implication of PROT_EXEC by PROT_READ */
1032 if (current->personality & READ_IMPLIES_EXEC) {
1033 if (capabilities & BDI_CAP_EXEC_MAP)
1034 prot |= PROT_EXEC;
1035 }
1036 }
1037 else if ((prot & PROT_READ) &&
1038 (prot & PROT_EXEC) &&
1039 !(capabilities & BDI_CAP_EXEC_MAP)
1040 ) {
1041 /* backing file is not executable, try to copy */
1042 capabilities &= ~BDI_CAP_MAP_DIRECT;
1043 }
1044 }
1045 else {
1046 /* anonymous mappings are always memory backed and can be
1047 * privately mapped
1048 */
1049 capabilities = BDI_CAP_MAP_COPY;
1050
1051 /* handle PROT_EXEC implication by PROT_READ */
1052 if ((prot & PROT_READ) &&
1053 (current->personality & READ_IMPLIES_EXEC))
1054 prot |= PROT_EXEC;
1055 }
1056
1057 /* allow the security API to have its say */
Eric Parised032182007-06-28 15:55:21 -04001058 ret = security_file_mmap(file, reqprot, prot, flags, addr, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 if (ret < 0)
1060 return ret;
1061
1062 /* looks okay */
1063 *_capabilities = capabilities;
1064 return 0;
1065}
1066
1067/*
1068 * we've determined that we can make the mapping, now translate what we
1069 * now know into VMA flags
1070 */
1071static unsigned long determine_vm_flags(struct file *file,
1072 unsigned long prot,
1073 unsigned long flags,
1074 unsigned long capabilities)
1075{
1076 unsigned long vm_flags;
1077
1078 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 /* vm_flags |= mm->def_flags; */
1080
1081 if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
1082 /* attempt to share read-only copies of mapped file chunks */
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001083 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 if (file && !(prot & PROT_WRITE))
1085 vm_flags |= VM_MAYSHARE;
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001086 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 /* overlay a shareable mapping on the backing device or inode
1088 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1089 * romfs/cramfs */
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001090 vm_flags |= VM_MAYSHARE | (capabilities & BDI_CAP_VMFLAGS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 if (flags & MAP_SHARED)
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001092 vm_flags |= VM_SHARED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 }
1094
1095 /* refuse to let anyone share private mappings with this process if
1096 * it's being traced - otherwise breakpoints set in it may interfere
1097 * with another untraced process
1098 */
Roland McGrathfa8e26c2008-07-25 19:45:50 -07001099 if ((flags & MAP_PRIVATE) && tracehook_expect_breakpoints(current))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 vm_flags &= ~VM_MAYSHARE;
1101
1102 return vm_flags;
1103}
1104
1105/*
David Howells8feae132009-01-08 12:04:47 +00001106 * set up a shared mapping on a file (the driver or filesystem provides and
1107 * pins the storage)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 */
David Howells8feae132009-01-08 12:04:47 +00001109static int do_mmap_shared_file(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110{
1111 int ret;
1112
1113 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001114 if (ret == 0) {
1115 vma->vm_region->vm_top = vma->vm_region->vm_end;
David Howells645d83c2009-09-24 15:13:10 +01001116 return 0;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001117 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118 if (ret != -ENOSYS)
1119 return ret;
1120
David Howells3fa30462010-03-23 13:35:21 -07001121 /* getting -ENOSYS indicates that direct mmap isn't possible (as
1122 * opposed to tried but failed) so we can only give a suitable error as
1123 * it's not possible to make a private copy if MAP_SHARED was given */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 return -ENODEV;
1125}
1126
1127/*
1128 * set up a private mapping or an anonymous shared mapping
1129 */
David Howells8feae132009-01-08 12:04:47 +00001130static int do_mmap_private(struct vm_area_struct *vma,
1131 struct vm_region *region,
David Howells645d83c2009-09-24 15:13:10 +01001132 unsigned long len,
1133 unsigned long capabilities)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134{
David Howells8feae132009-01-08 12:04:47 +00001135 struct page *pages;
1136 unsigned long total, point, n, rlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137 void *base;
David Howells8feae132009-01-08 12:04:47 +00001138 int ret, order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139
1140 /* invoke the file's mapping function so that it can keep track of
1141 * shared mappings on devices or memory
1142 * - VM_MAYSHARE will be set if it may attempt to share
1143 */
David Howells645d83c2009-09-24 15:13:10 +01001144 if (capabilities & BDI_CAP_MAP_DIRECT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001146 if (ret == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 /* shouldn't return success if we're not sharing */
Paul Mundtdd8632a2009-01-08 12:04:47 +00001148 BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1149 vma->vm_region->vm_top = vma->vm_region->vm_end;
David Howells645d83c2009-09-24 15:13:10 +01001150 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151 }
Paul Mundtdd8632a2009-01-08 12:04:47 +00001152 if (ret != -ENOSYS)
1153 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154
1155 /* getting an ENOSYS error indicates that direct mmap isn't
1156 * possible (as opposed to tried but failed) so we'll try to
1157 * make a private copy of the data and map that instead */
1158 }
1159
David Howells8feae132009-01-08 12:04:47 +00001160 rlen = PAGE_ALIGN(len);
1161
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162 /* allocate some memory to hold the mapping
1163 * - note that this may not return a page-aligned address if the object
1164 * we're allocating is smaller than a page
1165 */
David Howells8feae132009-01-08 12:04:47 +00001166 order = get_order(rlen);
1167 kdebug("alloc order %d for %lx", order, len);
1168
1169 pages = alloc_pages(GFP_KERNEL, order);
1170 if (!pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 goto enomem;
1172
David Howells8feae132009-01-08 12:04:47 +00001173 total = 1 << order;
David Howells33e5d7692009-04-02 16:56:32 -07001174 atomic_long_add(total, &mmap_pages_allocated);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175
David Howells8feae132009-01-08 12:04:47 +00001176 point = rlen >> PAGE_SHIFT;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001177
1178 /* we allocated a power-of-2 sized page set, so we may want to trim off
1179 * the excess */
1180 if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) {
1181 while (total > point) {
1182 order = ilog2(total - point);
1183 n = 1 << order;
1184 kdebug("shave %lu/%lu @%lu", n, total - point, total);
David Howells33e5d7692009-04-02 16:56:32 -07001185 atomic_long_sub(n, &mmap_pages_allocated);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001186 total -= n;
1187 set_page_refcounted(pages + total);
1188 __free_pages(pages + total, order);
1189 }
David Howells8feae132009-01-08 12:04:47 +00001190 }
1191
David Howells8feae132009-01-08 12:04:47 +00001192 for (point = 1; point < total; point++)
1193 set_page_refcounted(&pages[point]);
1194
1195 base = page_address(pages);
1196 region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1197 region->vm_start = (unsigned long) base;
1198 region->vm_end = region->vm_start + rlen;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001199 region->vm_top = region->vm_start + (total << PAGE_SHIFT);
David Howells8feae132009-01-08 12:04:47 +00001200
1201 vma->vm_start = region->vm_start;
1202 vma->vm_end = region->vm_start + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203
1204 if (vma->vm_file) {
1205 /* read the contents of a file into the copy */
1206 mm_segment_t old_fs;
1207 loff_t fpos;
1208
1209 fpos = vma->vm_pgoff;
1210 fpos <<= PAGE_SHIFT;
1211
1212 old_fs = get_fs();
1213 set_fs(KERNEL_DS);
David Howells8feae132009-01-08 12:04:47 +00001214 ret = vma->vm_file->f_op->read(vma->vm_file, base, rlen, &fpos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 set_fs(old_fs);
1216
1217 if (ret < 0)
1218 goto error_free;
1219
1220 /* clear the last little bit */
David Howells8feae132009-01-08 12:04:47 +00001221 if (ret < rlen)
1222 memset(base + ret, 0, rlen - ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 }
1225
1226 return 0;
1227
1228error_free:
David Howells8feae132009-01-08 12:04:47 +00001229 free_page_series(region->vm_start, region->vm_end);
1230 region->vm_start = vma->vm_start = 0;
1231 region->vm_end = vma->vm_end = 0;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001232 region->vm_top = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233 return ret;
1234
1235enomem:
Greg Ungerer05ae6fa2009-01-13 17:30:22 +10001236 printk("Allocation of length %lu from process %d (%s) failed\n",
1237 len, current->pid, current->comm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 show_free_areas();
1239 return -ENOMEM;
1240}
1241
1242/*
1243 * handle mapping creation for uClinux
1244 */
1245unsigned long do_mmap_pgoff(struct file *file,
1246 unsigned long addr,
1247 unsigned long len,
1248 unsigned long prot,
1249 unsigned long flags,
1250 unsigned long pgoff)
1251{
David Howells8feae132009-01-08 12:04:47 +00001252 struct vm_area_struct *vma;
1253 struct vm_region *region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 struct rb_node *rb;
David Howells8feae132009-01-08 12:04:47 +00001255 unsigned long capabilities, vm_flags, result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 int ret;
1257
David Howells8feae132009-01-08 12:04:47 +00001258 kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff);
1259
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260 /* decide whether we should attempt the mapping, and if so what sort of
1261 * mapping */
1262 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1263 &capabilities);
David Howells8feae132009-01-08 12:04:47 +00001264 if (ret < 0) {
1265 kleave(" = %d [val]", ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266 return ret;
David Howells8feae132009-01-08 12:04:47 +00001267 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268
David Howells06aab5a2009-09-24 12:33:48 +01001269 /* we ignore the address hint */
1270 addr = 0;
1271
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272 /* we've determined that we can make the mapping, now translate what we
1273 * now know into VMA flags */
1274 vm_flags = determine_vm_flags(file, prot, flags, capabilities);
1275
David Howells8feae132009-01-08 12:04:47 +00001276 /* we're going to need to record the mapping */
1277 region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1278 if (!region)
1279 goto error_getting_region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280
David Howells8feae132009-01-08 12:04:47 +00001281 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1282 if (!vma)
1283 goto error_getting_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284
David Howells1e2ae592010-01-15 17:01:33 -08001285 region->vm_usage = 1;
David Howells8feae132009-01-08 12:04:47 +00001286 region->vm_flags = vm_flags;
1287 region->vm_pgoff = pgoff;
1288
Rik van Riel5beb4932010-03-05 13:42:07 -08001289 INIT_LIST_HEAD(&vma->anon_vma_chain);
David Howells8feae132009-01-08 12:04:47 +00001290 vma->vm_flags = vm_flags;
1291 vma->vm_pgoff = pgoff;
1292
1293 if (file) {
1294 region->vm_file = file;
1295 get_file(file);
1296 vma->vm_file = file;
1297 get_file(file);
1298 if (vm_flags & VM_EXECUTABLE) {
1299 added_exe_file_vma(current->mm);
1300 vma->vm_mm = current->mm;
1301 }
1302 }
1303
1304 down_write(&nommu_region_sem);
1305
1306 /* if we want to share, we need to check for regions created by other
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 * mmap() calls that overlap with our proposed mapping
David Howells8feae132009-01-08 12:04:47 +00001308 * - we can only share with a superset match on most regular files
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 * - shared mappings on character devices and memory backed files are
1310 * permitted to overlap inexactly as far as we are concerned for in
1311 * these cases, sharing is handled in the driver or filesystem rather
1312 * than here
1313 */
1314 if (vm_flags & VM_MAYSHARE) {
David Howells8feae132009-01-08 12:04:47 +00001315 struct vm_region *pregion;
1316 unsigned long pglen, rpglen, pgend, rpgend, start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317
David Howells8feae132009-01-08 12:04:47 +00001318 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1319 pgend = pgoff + pglen;
David Howells165b2392007-03-22 00:11:24 -08001320
David Howells8feae132009-01-08 12:04:47 +00001321 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1322 pregion = rb_entry(rb, struct vm_region, vm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323
David Howells8feae132009-01-08 12:04:47 +00001324 if (!(pregion->vm_flags & VM_MAYSHARE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 continue;
1326
1327 /* search for overlapping mappings on the same file */
David Howells8feae132009-01-08 12:04:47 +00001328 if (pregion->vm_file->f_path.dentry->d_inode !=
1329 file->f_path.dentry->d_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330 continue;
1331
David Howells8feae132009-01-08 12:04:47 +00001332 if (pregion->vm_pgoff >= pgend)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333 continue;
1334
David Howells8feae132009-01-08 12:04:47 +00001335 rpglen = pregion->vm_end - pregion->vm_start;
1336 rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1337 rpgend = pregion->vm_pgoff + rpglen;
1338 if (pgoff >= rpgend)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 continue;
1340
David Howells8feae132009-01-08 12:04:47 +00001341 /* handle inexactly overlapping matches between
1342 * mappings */
1343 if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1344 !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1345 /* new mapping is not a subset of the region */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 if (!(capabilities & BDI_CAP_MAP_DIRECT))
1347 goto sharing_violation;
1348 continue;
1349 }
1350
David Howells8feae132009-01-08 12:04:47 +00001351 /* we've found a region we can share */
David Howells1e2ae592010-01-15 17:01:33 -08001352 pregion->vm_usage++;
David Howells8feae132009-01-08 12:04:47 +00001353 vma->vm_region = pregion;
1354 start = pregion->vm_start;
1355 start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1356 vma->vm_start = start;
1357 vma->vm_end = start + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358
David Howells8feae132009-01-08 12:04:47 +00001359 if (pregion->vm_flags & VM_MAPPED_COPY) {
1360 kdebug("share copy");
1361 vma->vm_flags |= VM_MAPPED_COPY;
1362 } else {
1363 kdebug("share mmap");
1364 ret = do_mmap_shared_file(vma);
1365 if (ret < 0) {
1366 vma->vm_region = NULL;
1367 vma->vm_start = 0;
1368 vma->vm_end = 0;
David Howells1e2ae592010-01-15 17:01:33 -08001369 pregion->vm_usage--;
David Howells8feae132009-01-08 12:04:47 +00001370 pregion = NULL;
1371 goto error_just_free;
1372 }
1373 }
1374 fput(region->vm_file);
1375 kmem_cache_free(vm_region_jar, region);
1376 region = pregion;
1377 result = start;
1378 goto share;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 }
1380
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 /* obtain the address at which to make a shared mapping
1382 * - this is the hook for quasi-memory character devices to
1383 * tell us the location of a shared mapping
1384 */
David Howells645d83c2009-09-24 15:13:10 +01001385 if (capabilities & BDI_CAP_MAP_DIRECT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 addr = file->f_op->get_unmapped_area(file, addr, len,
1387 pgoff, flags);
1388 if (IS_ERR((void *) addr)) {
1389 ret = addr;
1390 if (ret != (unsigned long) -ENOSYS)
David Howells8feae132009-01-08 12:04:47 +00001391 goto error_just_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392
1393 /* the driver refused to tell us where to site
1394 * the mapping so we'll have to attempt to copy
1395 * it */
1396 ret = (unsigned long) -ENODEV;
1397 if (!(capabilities & BDI_CAP_MAP_COPY))
David Howells8feae132009-01-08 12:04:47 +00001398 goto error_just_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399
1400 capabilities &= ~BDI_CAP_MAP_DIRECT;
David Howells8feae132009-01-08 12:04:47 +00001401 } else {
1402 vma->vm_start = region->vm_start = addr;
1403 vma->vm_end = region->vm_end = addr + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 }
1405 }
1406 }
1407
David Howells8feae132009-01-08 12:04:47 +00001408 vma->vm_region = region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409
David Howells645d83c2009-09-24 15:13:10 +01001410 /* set up the mapping
1411 * - the region is filled in if BDI_CAP_MAP_DIRECT is still set
1412 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 if (file && vma->vm_flags & VM_SHARED)
David Howells8feae132009-01-08 12:04:47 +00001414 ret = do_mmap_shared_file(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415 else
David Howells645d83c2009-09-24 15:13:10 +01001416 ret = do_mmap_private(vma, region, len, capabilities);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 if (ret < 0)
David Howells645d83c2009-09-24 15:13:10 +01001418 goto error_just_free;
1419 add_nommu_region(region);
David Howells8feae132009-01-08 12:04:47 +00001420
Jie Zhangea637632009-12-14 18:00:02 -08001421 /* clear anonymous mappings that don't ask for uninitialized data */
1422 if (!vma->vm_file && !(flags & MAP_UNINITIALIZED))
1423 memset((void *)region->vm_start, 0,
1424 region->vm_end - region->vm_start);
1425
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 /* okay... we have a mapping; now we have to register it */
David Howells8feae132009-01-08 12:04:47 +00001427 result = vma->vm_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429 current->mm->total_vm += len >> PAGE_SHIFT;
1430
David Howells8feae132009-01-08 12:04:47 +00001431share:
1432 add_vma_to_mm(current->mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433
Mike Frysingercfe79c02010-01-06 17:23:23 +00001434 /* we flush the region from the icache only when the first executable
1435 * mapping of it is made */
1436 if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1437 flush_icache_range(region->vm_start, region->vm_end);
1438 region->vm_icache_flushed = true;
1439 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440
Mike Frysingercfe79c02010-01-06 17:23:23 +00001441 up_write(&nommu_region_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442
David Howells8feae132009-01-08 12:04:47 +00001443 kleave(" = %lx", result);
1444 return result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445
David Howells8feae132009-01-08 12:04:47 +00001446error_just_free:
1447 up_write(&nommu_region_sem);
1448error:
David Howells89a86402009-10-30 13:13:26 +00001449 if (region->vm_file)
1450 fput(region->vm_file);
David Howells8feae132009-01-08 12:04:47 +00001451 kmem_cache_free(vm_region_jar, region);
David Howells89a86402009-10-30 13:13:26 +00001452 if (vma->vm_file)
1453 fput(vma->vm_file);
David Howells8feae132009-01-08 12:04:47 +00001454 if (vma->vm_flags & VM_EXECUTABLE)
1455 removed_exe_file_vma(vma->vm_mm);
1456 kmem_cache_free(vm_area_cachep, vma);
1457 kleave(" = %d", ret);
1458 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459
David Howells8feae132009-01-08 12:04:47 +00001460sharing_violation:
1461 up_write(&nommu_region_sem);
1462 printk(KERN_WARNING "Attempt to share mismatched mappings\n");
1463 ret = -EINVAL;
1464 goto error;
1465
1466error_getting_vma:
1467 kmem_cache_free(vm_region_jar, region);
1468 printk(KERN_WARNING "Allocation of vma for %lu byte allocation"
1469 " from process %d failed\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 len, current->pid);
1471 show_free_areas();
1472 return -ENOMEM;
1473
David Howells8feae132009-01-08 12:04:47 +00001474error_getting_region:
1475 printk(KERN_WARNING "Allocation of vm region for %lu byte allocation"
1476 " from process %d failed\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 len, current->pid);
1478 show_free_areas();
1479 return -ENOMEM;
1480}
Paul Mundtb5073172007-07-21 04:37:25 -07001481EXPORT_SYMBOL(do_mmap_pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001483SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1484 unsigned long, prot, unsigned long, flags,
1485 unsigned long, fd, unsigned long, pgoff)
1486{
1487 struct file *file = NULL;
1488 unsigned long retval = -EBADF;
1489
Al Viro120a7952010-10-30 02:54:44 -04001490 audit_mmap_fd(fd, flags);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001491 if (!(flags & MAP_ANONYMOUS)) {
1492 file = fget(fd);
1493 if (!file)
1494 goto out;
1495 }
1496
1497 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1498
1499 down_write(&current->mm->mmap_sem);
1500 retval = do_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1501 up_write(&current->mm->mmap_sem);
1502
1503 if (file)
1504 fput(file);
1505out:
1506 return retval;
1507}
1508
Christoph Hellwiga4679372010-03-10 15:21:15 -08001509#ifdef __ARCH_WANT_SYS_OLD_MMAP
1510struct mmap_arg_struct {
1511 unsigned long addr;
1512 unsigned long len;
1513 unsigned long prot;
1514 unsigned long flags;
1515 unsigned long fd;
1516 unsigned long offset;
1517};
1518
1519SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1520{
1521 struct mmap_arg_struct a;
1522
1523 if (copy_from_user(&a, arg, sizeof(a)))
1524 return -EFAULT;
1525 if (a.offset & ~PAGE_MASK)
1526 return -EINVAL;
1527
1528 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1529 a.offset >> PAGE_SHIFT);
1530}
1531#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1532
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533/*
David Howells8feae132009-01-08 12:04:47 +00001534 * split a vma into two pieces at address 'addr', a new vma is allocated either
1535 * for the first part or the tail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536 */
David Howells8feae132009-01-08 12:04:47 +00001537int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1538 unsigned long addr, int new_below)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539{
David Howells8feae132009-01-08 12:04:47 +00001540 struct vm_area_struct *new;
1541 struct vm_region *region;
1542 unsigned long npages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543
David Howells8feae132009-01-08 12:04:47 +00001544 kenter("");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545
David Howells779c1022010-01-15 17:01:34 -08001546 /* we're only permitted to split anonymous regions (these should have
1547 * only a single usage on the region) */
1548 if (vma->vm_file)
David Howells8feae132009-01-08 12:04:47 +00001549 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550
David Howells8feae132009-01-08 12:04:47 +00001551 if (mm->map_count >= sysctl_max_map_count)
1552 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553
David Howells8feae132009-01-08 12:04:47 +00001554 region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1555 if (!region)
1556 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557
David Howells8feae132009-01-08 12:04:47 +00001558 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1559 if (!new) {
1560 kmem_cache_free(vm_region_jar, region);
1561 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 }
David Howells8feae132009-01-08 12:04:47 +00001563
1564 /* most fields are the same, copy all, and then fixup */
1565 *new = *vma;
1566 *region = *vma->vm_region;
1567 new->vm_region = region;
1568
1569 npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1570
1571 if (new_below) {
Paul Mundtdd8632a2009-01-08 12:04:47 +00001572 region->vm_top = region->vm_end = new->vm_end = addr;
David Howells8feae132009-01-08 12:04:47 +00001573 } else {
1574 region->vm_start = new->vm_start = addr;
1575 region->vm_pgoff = new->vm_pgoff += npages;
1576 }
1577
1578 if (new->vm_ops && new->vm_ops->open)
1579 new->vm_ops->open(new);
1580
1581 delete_vma_from_mm(vma);
1582 down_write(&nommu_region_sem);
1583 delete_nommu_region(vma->vm_region);
1584 if (new_below) {
1585 vma->vm_region->vm_start = vma->vm_start = addr;
1586 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1587 } else {
1588 vma->vm_region->vm_end = vma->vm_end = addr;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001589 vma->vm_region->vm_top = addr;
David Howells8feae132009-01-08 12:04:47 +00001590 }
1591 add_nommu_region(vma->vm_region);
1592 add_nommu_region(new->vm_region);
1593 up_write(&nommu_region_sem);
1594 add_vma_to_mm(mm, vma);
1595 add_vma_to_mm(mm, new);
1596 return 0;
1597}
1598
1599/*
1600 * shrink a VMA by removing the specified chunk from either the beginning or
1601 * the end
1602 */
1603static int shrink_vma(struct mm_struct *mm,
1604 struct vm_area_struct *vma,
1605 unsigned long from, unsigned long to)
1606{
1607 struct vm_region *region;
1608
1609 kenter("");
1610
1611 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1612 * and list */
1613 delete_vma_from_mm(vma);
1614 if (from > vma->vm_start)
1615 vma->vm_end = from;
1616 else
1617 vma->vm_start = to;
1618 add_vma_to_mm(mm, vma);
1619
1620 /* cut the backing region down to size */
1621 region = vma->vm_region;
David Howells1e2ae592010-01-15 17:01:33 -08001622 BUG_ON(region->vm_usage != 1);
David Howells8feae132009-01-08 12:04:47 +00001623
1624 down_write(&nommu_region_sem);
1625 delete_nommu_region(region);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001626 if (from > region->vm_start) {
1627 to = region->vm_top;
1628 region->vm_top = region->vm_end = from;
1629 } else {
David Howells8feae132009-01-08 12:04:47 +00001630 region->vm_start = to;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001631 }
David Howells8feae132009-01-08 12:04:47 +00001632 add_nommu_region(region);
1633 up_write(&nommu_region_sem);
1634
1635 free_page_series(from, to);
1636 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637}
1638
David Howells30340972006-09-27 01:50:20 -07001639/*
1640 * release a mapping
David Howells8feae132009-01-08 12:04:47 +00001641 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1642 * VMA, though it need not cover the whole VMA
David Howells30340972006-09-27 01:50:20 -07001643 */
David Howells8feae132009-01-08 12:04:47 +00001644int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645{
David Howells8feae132009-01-08 12:04:47 +00001646 struct vm_area_struct *vma;
1647 struct rb_node *rb;
1648 unsigned long end = start + len;
1649 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650
David Howells8feae132009-01-08 12:04:47 +00001651 kenter(",%lx,%zx", start, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652
David Howells8feae132009-01-08 12:04:47 +00001653 if (len == 0)
1654 return -EINVAL;
1655
1656 /* find the first potentially overlapping VMA */
1657 vma = find_vma(mm, start);
1658 if (!vma) {
David Howells33e5d7692009-04-02 16:56:32 -07001659 static int limit = 0;
1660 if (limit < 5) {
1661 printk(KERN_WARNING
1662 "munmap of memory not mmapped by process %d"
1663 " (%s): 0x%lx-0x%lx\n",
1664 current->pid, current->comm,
1665 start, start + len - 1);
1666 limit++;
1667 }
David Howells8feae132009-01-08 12:04:47 +00001668 return -EINVAL;
David Howells30340972006-09-27 01:50:20 -07001669 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670
David Howells8feae132009-01-08 12:04:47 +00001671 /* we're allowed to split an anonymous VMA but not a file-backed one */
1672 if (vma->vm_file) {
1673 do {
1674 if (start > vma->vm_start) {
1675 kleave(" = -EINVAL [miss]");
1676 return -EINVAL;
1677 }
1678 if (end == vma->vm_end)
1679 goto erase_whole_vma;
1680 rb = rb_next(&vma->vm_rb);
1681 vma = rb_entry(rb, struct vm_area_struct, vm_rb);
1682 } while (rb);
1683 kleave(" = -EINVAL [split file]");
1684 return -EINVAL;
1685 } else {
1686 /* the chunk must be a subset of the VMA found */
1687 if (start == vma->vm_start && end == vma->vm_end)
1688 goto erase_whole_vma;
1689 if (start < vma->vm_start || end > vma->vm_end) {
1690 kleave(" = -EINVAL [superset]");
1691 return -EINVAL;
1692 }
1693 if (start & ~PAGE_MASK) {
1694 kleave(" = -EINVAL [unaligned start]");
1695 return -EINVAL;
1696 }
1697 if (end != vma->vm_end && end & ~PAGE_MASK) {
1698 kleave(" = -EINVAL [unaligned split]");
1699 return -EINVAL;
1700 }
1701 if (start != vma->vm_start && end != vma->vm_end) {
1702 ret = split_vma(mm, vma, start, 1);
1703 if (ret < 0) {
1704 kleave(" = %d [split]", ret);
1705 return ret;
1706 }
1707 }
1708 return shrink_vma(mm, vma, start, end);
1709 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710
David Howells8feae132009-01-08 12:04:47 +00001711erase_whole_vma:
1712 delete_vma_from_mm(vma);
1713 delete_vma(mm, vma);
1714 kleave(" = 0");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 return 0;
1716}
Paul Mundtb5073172007-07-21 04:37:25 -07001717EXPORT_SYMBOL(do_munmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718
Heiko Carstens6a6160a2009-01-14 14:14:15 +01001719SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
David Howells30340972006-09-27 01:50:20 -07001720{
1721 int ret;
1722 struct mm_struct *mm = current->mm;
1723
1724 down_write(&mm->mmap_sem);
1725 ret = do_munmap(mm, addr, len);
1726 up_write(&mm->mmap_sem);
1727 return ret;
1728}
1729
1730/*
David Howells8feae132009-01-08 12:04:47 +00001731 * release all the mappings made in a process's VM space
David Howells30340972006-09-27 01:50:20 -07001732 */
David Howells8feae132009-01-08 12:04:47 +00001733void exit_mmap(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734{
David Howells8feae132009-01-08 12:04:47 +00001735 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736
David Howells8feae132009-01-08 12:04:47 +00001737 if (!mm)
1738 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739
David Howells8feae132009-01-08 12:04:47 +00001740 kenter("");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741
David Howells8feae132009-01-08 12:04:47 +00001742 mm->total_vm = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743
David Howells8feae132009-01-08 12:04:47 +00001744 while ((vma = mm->mmap)) {
1745 mm->mmap = vma->vm_next;
1746 delete_vma_from_mm(vma);
1747 delete_vma(mm, vma);
Steven J. Magnani04c34962010-11-24 12:56:54 -08001748 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749 }
David Howells8feae132009-01-08 12:04:47 +00001750
1751 kleave("");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752}
1753
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754unsigned long do_brk(unsigned long addr, unsigned long len)
1755{
1756 return -ENOMEM;
1757}
1758
1759/*
David Howells6fa5f802006-09-27 01:50:21 -07001760 * expand (or shrink) an existing mapping, potentially moving it at the same
1761 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 *
David Howells6fa5f802006-09-27 01:50:21 -07001763 * under NOMMU conditions, we only permit changing a mapping's size, and only
David Howells8feae132009-01-08 12:04:47 +00001764 * as long as it stays within the region allocated by do_mmap_private() and the
1765 * block is not shareable
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 *
David Howells6fa5f802006-09-27 01:50:21 -07001767 * MREMAP_FIXED is not supported under NOMMU conditions
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 */
1769unsigned long do_mremap(unsigned long addr,
1770 unsigned long old_len, unsigned long new_len,
1771 unsigned long flags, unsigned long new_addr)
1772{
David Howells6fa5f802006-09-27 01:50:21 -07001773 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774
1775 /* insanity checks first */
David Howells8feae132009-01-08 12:04:47 +00001776 if (old_len == 0 || new_len == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777 return (unsigned long) -EINVAL;
1778
David Howells8feae132009-01-08 12:04:47 +00001779 if (addr & ~PAGE_MASK)
1780 return -EINVAL;
1781
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782 if (flags & MREMAP_FIXED && new_addr != addr)
1783 return (unsigned long) -EINVAL;
1784
David Howells8feae132009-01-08 12:04:47 +00001785 vma = find_vma_exact(current->mm, addr, old_len);
David Howells6fa5f802006-09-27 01:50:21 -07001786 if (!vma)
1787 return (unsigned long) -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788
David Howells6fa5f802006-09-27 01:50:21 -07001789 if (vma->vm_end != vma->vm_start + old_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790 return (unsigned long) -EFAULT;
1791
David Howells6fa5f802006-09-27 01:50:21 -07001792 if (vma->vm_flags & VM_MAYSHARE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793 return (unsigned long) -EPERM;
1794
David Howells8feae132009-01-08 12:04:47 +00001795 if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796 return (unsigned long) -ENOMEM;
1797
1798 /* all checks complete - do it */
David Howells6fa5f802006-09-27 01:50:21 -07001799 vma->vm_end = vma->vm_start + new_len;
David Howells6fa5f802006-09-27 01:50:21 -07001800 return vma->vm_start;
1801}
Paul Mundtb5073172007-07-21 04:37:25 -07001802EXPORT_SYMBOL(do_mremap);
David Howells6fa5f802006-09-27 01:50:21 -07001803
Heiko Carstens6a6160a2009-01-14 14:14:15 +01001804SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1805 unsigned long, new_len, unsigned long, flags,
1806 unsigned long, new_addr)
David Howells6fa5f802006-09-27 01:50:21 -07001807{
1808 unsigned long ret;
1809
1810 down_write(&current->mm->mmap_sem);
1811 ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1812 up_write(&current->mm->mmap_sem);
1813 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814}
1815
Linus Torvalds6aab3412005-11-28 14:34:23 -08001816struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001817 unsigned int foll_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818{
1819 return NULL;
1820}
1821
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822int remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
1823 unsigned long to, unsigned long size, pgprot_t prot)
1824{
Greg Ungerer66aa2b42005-09-12 11:18:10 +10001825 vma->vm_start = vma->vm_pgoff << PAGE_SHIFT;
1826 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827}
Luke Yang22c4af42006-07-14 00:24:09 -07001828EXPORT_SYMBOL(remap_pfn_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829
Paul Mundtf905bc42008-02-04 22:29:59 -08001830int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1831 unsigned long pgoff)
1832{
1833 unsigned int size = vma->vm_end - vma->vm_start;
1834
1835 if (!(vma->vm_flags & VM_USERMAP))
1836 return -EINVAL;
1837
1838 vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1839 vma->vm_end = vma->vm_start + size;
1840
1841 return 0;
1842}
1843EXPORT_SYMBOL(remap_vmalloc_range);
1844
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1846 unsigned long len, unsigned long pgoff, unsigned long flags)
1847{
1848 return -ENOMEM;
1849}
1850
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001851void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852{
1853}
1854
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855void unmap_mapping_range(struct address_space *mapping,
1856 loff_t const holebegin, loff_t const holelen,
1857 int even_cows)
1858{
1859}
Luke Yang22c4af42006-07-14 00:24:09 -07001860EXPORT_SYMBOL(unmap_mapping_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861
1862/*
1863 * Check that a process has enough memory to allocate a new virtual
1864 * mapping. 0 means there is enough memory for the allocation to
1865 * succeed and -ENOMEM implies there is not.
1866 *
1867 * We currently support three overcommit policies, which are set via the
1868 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
1869 *
1870 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1871 * Additional code 2002 Jul 20 by Robert Love.
1872 *
1873 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1874 *
1875 * Note this is a helper function intended to be used by LSMs which
1876 * wish to use this logic.
1877 */
Alan Cox34b4e4a2007-08-22 14:01:28 -07001878int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879{
1880 unsigned long free, allowed;
1881
1882 vm_acct_memory(pages);
1883
1884 /*
1885 * Sometimes we want to use more memory than we have
1886 */
1887 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1888 return 0;
1889
1890 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1891 unsigned long n;
1892
Christoph Lameter347ce432006-06-30 01:55:35 -07001893 free = global_page_state(NR_FILE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894 free += nr_swap_pages;
1895
1896 /*
1897 * Any slabs which are created with the
1898 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1899 * which are reclaimable, under pressure. The dentry
1900 * cache and most inode caches should fall into this
1901 */
Christoph Lameter972d1a72006-09-25 23:31:51 -07001902 free += global_page_state(NR_SLAB_RECLAIMABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903
1904 /*
1905 * Leave the last 3% for root
1906 */
1907 if (!cap_sys_admin)
1908 free -= free / 32;
1909
1910 if (free > pages)
1911 return 0;
1912
1913 /*
1914 * nr_free_pages() is very expensive on large systems,
1915 * only call if we're about to fail.
1916 */
1917 n = nr_free_pages();
Hideo AOKId5ddc792006-04-10 22:53:01 -07001918
1919 /*
1920 * Leave reserved pages. The pages are not for anonymous pages.
1921 */
1922 if (n <= totalreserve_pages)
1923 goto error;
1924 else
1925 n -= totalreserve_pages;
1926
1927 /*
1928 * Leave the last 3% for root
1929 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930 if (!cap_sys_admin)
1931 n -= n / 32;
1932 free += n;
1933
1934 if (free > pages)
1935 return 0;
Hideo AOKId5ddc792006-04-10 22:53:01 -07001936
1937 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938 }
1939
1940 allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1941 /*
1942 * Leave the last 3% for root
1943 */
1944 if (!cap_sys_admin)
1945 allowed -= allowed / 32;
1946 allowed += total_swap_pages;
1947
1948 /* Don't let a single process grow too big:
1949 leave 3% of the size of this process for other processes */
Alan Cox731572d2008-10-29 14:01:20 -07001950 if (mm)
1951 allowed -= mm->total_vm / 32;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07001953 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954 return 0;
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07001955
Hideo AOKId5ddc792006-04-10 22:53:01 -07001956error:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957 vm_unacct_memory(pages);
1958
1959 return -ENOMEM;
1960}
1961
Stephen Wilsoncae5d392011-03-13 15:49:17 -04001962int in_gate_area_no_mm(unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963{
1964 return 0;
1965}
David Howellsb0e15192006-01-06 00:11:42 -08001966
Nick Piggind0217ac2007-07-19 01:47:03 -07001967int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
David Howellsb0e15192006-01-06 00:11:42 -08001968{
1969 BUG();
Nick Piggind0217ac2007-07-19 01:47:03 -07001970 return 0;
David Howellsb0e15192006-01-06 00:11:42 -08001971}
Paul Mundtb5073172007-07-21 04:37:25 -07001972EXPORT_SYMBOL(filemap_fault);
David Howells0ec76a12006-09-27 01:50:15 -07001973
Mike Frysingerf55f1992011-03-29 14:05:12 +01001974static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
1975 unsigned long addr, void *buf, int len, int write)
David Howells0ec76a12006-09-27 01:50:15 -07001976{
David Howells0ec76a12006-09-27 01:50:15 -07001977 struct vm_area_struct *vma;
David Howells0ec76a12006-09-27 01:50:15 -07001978
1979 down_read(&mm->mmap_sem);
1980
1981 /* the access must start within one of the target process's mappings */
David Howells0159b142006-09-27 01:50:16 -07001982 vma = find_vma(mm, addr);
1983 if (vma) {
David Howells0ec76a12006-09-27 01:50:15 -07001984 /* don't overrun this mapping */
1985 if (addr + len >= vma->vm_end)
1986 len = vma->vm_end - addr;
1987
1988 /* only read or write mappings where it is permitted */
David Howellsd00c7b992006-09-27 01:50:19 -07001989 if (write && vma->vm_flags & VM_MAYWRITE)
Jie Zhang79597222010-01-06 17:23:28 +00001990 copy_to_user_page(vma, NULL, addr,
1991 (void *) addr, buf, len);
David Howellsd00c7b992006-09-27 01:50:19 -07001992 else if (!write && vma->vm_flags & VM_MAYREAD)
Jie Zhang79597222010-01-06 17:23:28 +00001993 copy_from_user_page(vma, NULL, addr,
1994 buf, (void *) addr, len);
David Howells0ec76a12006-09-27 01:50:15 -07001995 else
1996 len = 0;
1997 } else {
1998 len = 0;
1999 }
2000
2001 up_read(&mm->mmap_sem);
Mike Frysingerf55f1992011-03-29 14:05:12 +01002002
2003 return len;
2004}
2005
2006/**
2007 * @access_remote_vm - access another process' address space
2008 * @mm: the mm_struct of the target address space
2009 * @addr: start address to access
2010 * @buf: source or destination buffer
2011 * @len: number of bytes to transfer
2012 * @write: whether the access is a write
2013 *
2014 * The caller must hold a reference on @mm.
2015 */
2016int access_remote_vm(struct mm_struct *mm, unsigned long addr,
2017 void *buf, int len, int write)
2018{
2019 return __access_remote_vm(NULL, mm, addr, buf, len, write);
2020}
2021
2022/*
2023 * Access another process' address space.
2024 * - source/target buffer must be kernel space
2025 */
2026int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
2027{
2028 struct mm_struct *mm;
2029
2030 if (addr + len < addr)
2031 return 0;
2032
2033 mm = get_task_mm(tsk);
2034 if (!mm)
2035 return 0;
2036
2037 len = __access_remote_vm(tsk, mm, addr, buf, len, write);
2038
David Howells0ec76a12006-09-27 01:50:15 -07002039 mmput(mm);
2040 return len;
2041}
David Howells7e660872010-01-15 17:01:39 -08002042
2043/**
2044 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
2045 * @inode: The inode to check
2046 * @size: The current filesize of the inode
2047 * @newsize: The proposed filesize of the inode
2048 *
2049 * Check the shared mappings on an inode on behalf of a shrinking truncate to
2050 * make sure that that any outstanding VMAs aren't broken and then shrink the
2051 * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
2052 * automatically grant mappings that are too large.
2053 */
2054int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
2055 size_t newsize)
2056{
2057 struct vm_area_struct *vma;
2058 struct prio_tree_iter iter;
2059 struct vm_region *region;
2060 pgoff_t low, high;
2061 size_t r_size, r_top;
2062
2063 low = newsize >> PAGE_SHIFT;
2064 high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
2065
2066 down_write(&nommu_region_sem);
2067
2068 /* search for VMAs that fall within the dead zone */
2069 vma_prio_tree_foreach(vma, &iter, &inode->i_mapping->i_mmap,
2070 low, high) {
2071 /* found one - only interested if it's shared out of the page
2072 * cache */
2073 if (vma->vm_flags & VM_SHARED) {
2074 up_write(&nommu_region_sem);
2075 return -ETXTBSY; /* not quite true, but near enough */
2076 }
2077 }
2078
2079 /* reduce any regions that overlap the dead zone - if in existence,
2080 * these will be pointed to by VMAs that don't overlap the dead zone
2081 *
2082 * we don't check for any regions that start beyond the EOF as there
2083 * shouldn't be any
2084 */
2085 vma_prio_tree_foreach(vma, &iter, &inode->i_mapping->i_mmap,
2086 0, ULONG_MAX) {
2087 if (!(vma->vm_flags & VM_SHARED))
2088 continue;
2089
2090 region = vma->vm_region;
2091 r_size = region->vm_top - region->vm_start;
2092 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
2093
2094 if (r_top > newsize) {
2095 region->vm_top -= r_top - newsize;
2096 if (region->vm_end > region->vm_top)
2097 region->vm_end = region->vm_top;
2098 }
2099 }
2100
2101 up_write(&nommu_region_sem);
2102 return 0;
2103}