blob: 0563fd9003df86b0c8c39844341bfc1847d4dba7 [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 Dobriyanf0f37e2f2009-09-27 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{
Namhyung Kim6038def2011-05-24 17:11:22 -0700683 struct vm_area_struct *pvma, *prev;
David Howells8feae132009-01-08 12:04:47 +0000684 struct address_space *mapping;
Namhyung Kim6038def2011-05-24 17:11:22 -0700685 struct rb_node **p, *parent, *rb_prev;
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 */
Namhyung Kim6038def2011-05-24 17:11:22 -0700706 parent = rb_prev = NULL;
David Howells8feae132009-01-08 12:04:47 +0000707 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;
Namhyung Kim6038def2011-05-24 17:11:22 -0700716 else if (vma->vm_start > pvma->vm_start) {
717 rb_prev = parent;
David Howells8feae132009-01-08 12:04:47 +0000718 p = &(*p)->rb_right;
Namhyung Kim6038def2011-05-24 17:11:22 -0700719 } else if (vma->vm_end < pvma->vm_end)
David Howells8feae132009-01-08 12:04:47 +0000720 p = &(*p)->rb_left;
Namhyung Kim6038def2011-05-24 17:11:22 -0700721 else if (vma->vm_end > pvma->vm_end) {
722 rb_prev = parent;
David Howells8feae132009-01-08 12:04:47 +0000723 p = &(*p)->rb_right;
Namhyung Kim6038def2011-05-24 17:11:22 -0700724 } else if (vma < pvma)
David Howells8feae132009-01-08 12:04:47 +0000725 p = &(*p)->rb_left;
Namhyung Kim6038def2011-05-24 17:11:22 -0700726 else if (vma > pvma) {
727 rb_prev = parent;
David Howells8feae132009-01-08 12:04:47 +0000728 p = &(*p)->rb_right;
Namhyung Kim6038def2011-05-24 17:11:22 -0700729 } else
David Howells8feae132009-01-08 12:04:47 +0000730 BUG();
731 }
732
733 rb_link_node(&vma->vm_rb, parent, p);
734 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
735
736 /* add VMA to the VMA list also */
Namhyung Kim6038def2011-05-24 17:11:22 -0700737 prev = NULL;
738 if (rb_prev)
739 prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
David Howells30340972006-09-27 01:50:20 -0700740
Namhyung Kim6038def2011-05-24 17:11:22 -0700741 __vma_link_list(mm, vma, prev, parent);
David Howells8feae132009-01-08 12:04:47 +0000742}
743
744/*
745 * delete a VMA from its owning mm_struct and address space
746 */
747static void delete_vma_from_mm(struct vm_area_struct *vma)
748{
David Howells8feae132009-01-08 12:04:47 +0000749 struct address_space *mapping;
750 struct mm_struct *mm = vma->vm_mm;
751
752 kenter("%p", vma);
753
Bernd Schmidteb8cdec2009-09-21 17:03:57 -0700754 protect_vma(vma, 0);
755
David Howells8feae132009-01-08 12:04:47 +0000756 mm->map_count--;
757 if (mm->mmap_cache == vma)
758 mm->mmap_cache = NULL;
759
760 /* remove the VMA from the mapping */
761 if (vma->vm_file) {
762 mapping = vma->vm_file->f_mapping;
763
764 flush_dcache_mmap_lock(mapping);
765 vma_prio_tree_remove(vma, &mapping->i_mmap);
766 flush_dcache_mmap_unlock(mapping);
767 }
768
769 /* remove from the MM's tree and list */
770 rb_erase(&vma->vm_rb, &mm->mm_rb);
Namhyung Kimb951bf22011-05-24 17:11:23 -0700771
772 if (vma->vm_prev)
773 vma->vm_prev->vm_next = vma->vm_next;
774 else
775 mm->mmap = vma->vm_next;
776
777 if (vma->vm_next)
778 vma->vm_next->vm_prev = vma->vm_prev;
David Howells8feae132009-01-08 12:04:47 +0000779
780 vma->vm_mm = NULL;
781}
782
783/*
784 * destroy a VMA record
785 */
786static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
787{
788 kenter("%p", vma);
789 if (vma->vm_ops && vma->vm_ops->close)
790 vma->vm_ops->close(vma);
791 if (vma->vm_file) {
792 fput(vma->vm_file);
793 if (vma->vm_flags & VM_EXECUTABLE)
794 removed_exe_file_vma(mm);
795 }
796 put_nommu_region(vma->vm_region);
797 kmem_cache_free(vm_area_cachep, vma);
David Howells30340972006-09-27 01:50:20 -0700798}
799
800/*
801 * look up the first VMA in which addr resides, NULL if none
802 * - should be called with mm->mmap_sem at least held readlocked
803 */
804struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
805{
David Howells8feae132009-01-08 12:04:47 +0000806 struct vm_area_struct *vma;
David Howells30340972006-09-27 01:50:20 -0700807
David Howells8feae132009-01-08 12:04:47 +0000808 /* check the cache first */
809 vma = mm->mmap_cache;
810 if (vma && vma->vm_start <= addr && vma->vm_end > addr)
811 return vma;
812
Namhyung Kime922c4c2011-05-24 17:11:24 -0700813 /* trawl the list (there may be multiple mappings in which addr
David Howells8feae132009-01-08 12:04:47 +0000814 * resides) */
Namhyung Kime922c4c2011-05-24 17:11:24 -0700815 for (vma = mm->mmap; vma; vma = vma->vm_next) {
David Howells8feae132009-01-08 12:04:47 +0000816 if (vma->vm_start > addr)
817 return NULL;
818 if (vma->vm_end > addr) {
819 mm->mmap_cache = vma;
820 return vma;
821 }
David Howells30340972006-09-27 01:50:20 -0700822 }
823
David Howells30340972006-09-27 01:50:20 -0700824 return NULL;
825}
826EXPORT_SYMBOL(find_vma);
827
828/*
David Howells930e6522006-09-27 01:50:22 -0700829 * find a VMA
830 * - we don't extend stack VMAs under NOMMU conditions
831 */
832struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
833{
David Howells7561e8c2010-03-25 16:48:38 +0000834 return find_vma(mm, addr);
David Howells930e6522006-09-27 01:50:22 -0700835}
836
David Howells8feae132009-01-08 12:04:47 +0000837/*
838 * expand a stack to a given address
839 * - not supported under NOMMU conditions
840 */
Greg Ungerer57c8f632007-07-15 23:38:28 -0700841int expand_stack(struct vm_area_struct *vma, unsigned long address)
842{
843 return -ENOMEM;
844}
845
David Howells930e6522006-09-27 01:50:22 -0700846/*
David Howells6fa5f802006-09-27 01:50:21 -0700847 * look up the first VMA exactly that exactly matches addr
848 * - should be called with mm->mmap_sem at least held readlocked
849 */
David Howells8feae132009-01-08 12:04:47 +0000850static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
851 unsigned long addr,
852 unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853{
854 struct vm_area_struct *vma;
David Howells8feae132009-01-08 12:04:47 +0000855 unsigned long end = addr + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856
David Howells8feae132009-01-08 12:04:47 +0000857 /* check the cache first */
858 vma = mm->mmap_cache;
859 if (vma && vma->vm_start == addr && vma->vm_end == end)
860 return vma;
861
Namhyung Kime922c4c2011-05-24 17:11:24 -0700862 /* trawl the list (there may be multiple mappings in which addr
David Howells8feae132009-01-08 12:04:47 +0000863 * resides) */
Namhyung Kime922c4c2011-05-24 17:11:24 -0700864 for (vma = mm->mmap; vma; vma = vma->vm_next) {
David Howells8feae132009-01-08 12:04:47 +0000865 if (vma->vm_start < addr)
866 continue;
867 if (vma->vm_start > addr)
868 return NULL;
869 if (vma->vm_end == end) {
870 mm->mmap_cache = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 return vma;
David Howells8feae132009-01-08 12:04:47 +0000872 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 }
874
875 return NULL;
876}
877
David Howells30340972006-09-27 01:50:20 -0700878/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 * determine whether a mapping should be permitted and, if so, what sort of
880 * mapping we're capable of supporting
881 */
882static int validate_mmap_request(struct file *file,
883 unsigned long addr,
884 unsigned long len,
885 unsigned long prot,
886 unsigned long flags,
887 unsigned long pgoff,
888 unsigned long *_capabilities)
889{
David Howells8feae132009-01-08 12:04:47 +0000890 unsigned long capabilities, rlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 unsigned long reqprot = prot;
892 int ret;
893
894 /* do the simple checks first */
David Howells06aab5a2009-09-24 12:33:48 +0100895 if (flags & MAP_FIXED) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896 printk(KERN_DEBUG
897 "%d: Can't do fixed-address/overlay mmap of RAM\n",
898 current->pid);
899 return -EINVAL;
900 }
901
902 if ((flags & MAP_TYPE) != MAP_PRIVATE &&
903 (flags & MAP_TYPE) != MAP_SHARED)
904 return -EINVAL;
905
Mike Frysingerf81cff0d2006-12-06 12:02:59 +1000906 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 return -EINVAL;
908
Mike Frysingerf81cff0d2006-12-06 12:02:59 +1000909 /* Careful about overflows.. */
David Howells8feae132009-01-08 12:04:47 +0000910 rlen = PAGE_ALIGN(len);
911 if (!rlen || rlen > TASK_SIZE)
Mike Frysingerf81cff0d2006-12-06 12:02:59 +1000912 return -ENOMEM;
913
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 /* offset overflow? */
David Howells8feae132009-01-08 12:04:47 +0000915 if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
Mike Frysingerf81cff0d2006-12-06 12:02:59 +1000916 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917
918 if (file) {
919 /* validate file mapping requests */
920 struct address_space *mapping;
921
922 /* files must support mmap */
923 if (!file->f_op || !file->f_op->mmap)
924 return -ENODEV;
925
926 /* work out if what we've got could possibly be shared
927 * - we support chardevs that provide their own "memory"
928 * - we support files/blockdevs that are memory backed
929 */
930 mapping = file->f_mapping;
931 if (!mapping)
Josef Sipeke9536ae2006-12-08 02:37:21 -0800932 mapping = file->f_path.dentry->d_inode->i_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933
934 capabilities = 0;
935 if (mapping && mapping->backing_dev_info)
936 capabilities = mapping->backing_dev_info->capabilities;
937
938 if (!capabilities) {
939 /* no explicit capabilities set, so assume some
940 * defaults */
Josef Sipeke9536ae2006-12-08 02:37:21 -0800941 switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 case S_IFREG:
943 case S_IFBLK:
944 capabilities = BDI_CAP_MAP_COPY;
945 break;
946
947 case S_IFCHR:
948 capabilities =
949 BDI_CAP_MAP_DIRECT |
950 BDI_CAP_READ_MAP |
951 BDI_CAP_WRITE_MAP;
952 break;
953
954 default:
955 return -EINVAL;
956 }
957 }
958
959 /* eliminate any capabilities that we can't support on this
960 * device */
961 if (!file->f_op->get_unmapped_area)
962 capabilities &= ~BDI_CAP_MAP_DIRECT;
963 if (!file->f_op->read)
964 capabilities &= ~BDI_CAP_MAP_COPY;
965
Graff Yang28d7a6a2009-08-18 14:11:17 -0700966 /* The file shall have been opened with read permission. */
967 if (!(file->f_mode & FMODE_READ))
968 return -EACCES;
969
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 if (flags & MAP_SHARED) {
971 /* do checks for writing, appending and locking */
972 if ((prot & PROT_WRITE) &&
973 !(file->f_mode & FMODE_WRITE))
974 return -EACCES;
975
Josef Sipeke9536ae2006-12-08 02:37:21 -0800976 if (IS_APPEND(file->f_path.dentry->d_inode) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 (file->f_mode & FMODE_WRITE))
978 return -EACCES;
979
Josef Sipeke9536ae2006-12-08 02:37:21 -0800980 if (locks_verify_locked(file->f_path.dentry->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 return -EAGAIN;
982
983 if (!(capabilities & BDI_CAP_MAP_DIRECT))
984 return -ENODEV;
985
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 /* we mustn't privatise shared mappings */
987 capabilities &= ~BDI_CAP_MAP_COPY;
988 }
989 else {
990 /* we're going to read the file into private memory we
991 * allocate */
992 if (!(capabilities & BDI_CAP_MAP_COPY))
993 return -ENODEV;
994
995 /* we don't permit a private writable mapping to be
996 * shared with the backing device */
997 if (prot & PROT_WRITE)
998 capabilities &= ~BDI_CAP_MAP_DIRECT;
999 }
1000
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001001 if (capabilities & BDI_CAP_MAP_DIRECT) {
1002 if (((prot & PROT_READ) && !(capabilities & BDI_CAP_READ_MAP)) ||
1003 ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
1004 ((prot & PROT_EXEC) && !(capabilities & BDI_CAP_EXEC_MAP))
1005 ) {
1006 capabilities &= ~BDI_CAP_MAP_DIRECT;
1007 if (flags & MAP_SHARED) {
1008 printk(KERN_WARNING
1009 "MAP_SHARED not completely supported on !MMU\n");
1010 return -EINVAL;
1011 }
1012 }
1013 }
1014
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 /* handle executable mappings and implied executable
1016 * mappings */
Josef Sipeke9536ae2006-12-08 02:37:21 -08001017 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 if (prot & PROT_EXEC)
1019 return -EPERM;
1020 }
1021 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
1022 /* handle implication of PROT_EXEC by PROT_READ */
1023 if (current->personality & READ_IMPLIES_EXEC) {
1024 if (capabilities & BDI_CAP_EXEC_MAP)
1025 prot |= PROT_EXEC;
1026 }
1027 }
1028 else if ((prot & PROT_READ) &&
1029 (prot & PROT_EXEC) &&
1030 !(capabilities & BDI_CAP_EXEC_MAP)
1031 ) {
1032 /* backing file is not executable, try to copy */
1033 capabilities &= ~BDI_CAP_MAP_DIRECT;
1034 }
1035 }
1036 else {
1037 /* anonymous mappings are always memory backed and can be
1038 * privately mapped
1039 */
1040 capabilities = BDI_CAP_MAP_COPY;
1041
1042 /* handle PROT_EXEC implication by PROT_READ */
1043 if ((prot & PROT_READ) &&
1044 (current->personality & READ_IMPLIES_EXEC))
1045 prot |= PROT_EXEC;
1046 }
1047
1048 /* allow the security API to have its say */
Eric Parised032182007-06-28 15:55:21 -04001049 ret = security_file_mmap(file, reqprot, prot, flags, addr, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 if (ret < 0)
1051 return ret;
1052
1053 /* looks okay */
1054 *_capabilities = capabilities;
1055 return 0;
1056}
1057
1058/*
1059 * we've determined that we can make the mapping, now translate what we
1060 * now know into VMA flags
1061 */
1062static unsigned long determine_vm_flags(struct file *file,
1063 unsigned long prot,
1064 unsigned long flags,
1065 unsigned long capabilities)
1066{
1067 unsigned long vm_flags;
1068
1069 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 /* vm_flags |= mm->def_flags; */
1071
1072 if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
1073 /* attempt to share read-only copies of mapped file chunks */
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001074 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 if (file && !(prot & PROT_WRITE))
1076 vm_flags |= VM_MAYSHARE;
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001077 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 /* overlay a shareable mapping on the backing device or inode
1079 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1080 * romfs/cramfs */
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001081 vm_flags |= VM_MAYSHARE | (capabilities & BDI_CAP_VMFLAGS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 if (flags & MAP_SHARED)
Bernd Schmidt3c7b2042010-05-25 23:43:00 -07001083 vm_flags |= VM_SHARED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 }
1085
1086 /* refuse to let anyone share private mappings with this process if
1087 * it's being traced - otherwise breakpoints set in it may interfere
1088 * with another untraced process
1089 */
Roland McGrathfa8e26c2008-07-25 19:45:50 -07001090 if ((flags & MAP_PRIVATE) && tracehook_expect_breakpoints(current))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 vm_flags &= ~VM_MAYSHARE;
1092
1093 return vm_flags;
1094}
1095
1096/*
David Howells8feae132009-01-08 12:04:47 +00001097 * set up a shared mapping on a file (the driver or filesystem provides and
1098 * pins the storage)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 */
David Howells8feae132009-01-08 12:04:47 +00001100static int do_mmap_shared_file(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101{
1102 int ret;
1103
1104 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001105 if (ret == 0) {
1106 vma->vm_region->vm_top = vma->vm_region->vm_end;
David Howells645d83c2009-09-24 15:13:10 +01001107 return 0;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001108 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109 if (ret != -ENOSYS)
1110 return ret;
1111
David Howells3fa30462010-03-23 13:35:21 -07001112 /* getting -ENOSYS indicates that direct mmap isn't possible (as
1113 * opposed to tried but failed) so we can only give a suitable error as
1114 * it's not possible to make a private copy if MAP_SHARED was given */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 return -ENODEV;
1116}
1117
1118/*
1119 * set up a private mapping or an anonymous shared mapping
1120 */
David Howells8feae132009-01-08 12:04:47 +00001121static int do_mmap_private(struct vm_area_struct *vma,
1122 struct vm_region *region,
David Howells645d83c2009-09-24 15:13:10 +01001123 unsigned long len,
1124 unsigned long capabilities)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125{
David Howells8feae132009-01-08 12:04:47 +00001126 struct page *pages;
1127 unsigned long total, point, n, rlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 void *base;
David Howells8feae132009-01-08 12:04:47 +00001129 int ret, order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130
1131 /* invoke the file's mapping function so that it can keep track of
1132 * shared mappings on devices or memory
1133 * - VM_MAYSHARE will be set if it may attempt to share
1134 */
David Howells645d83c2009-09-24 15:13:10 +01001135 if (capabilities & BDI_CAP_MAP_DIRECT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001137 if (ret == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138 /* shouldn't return success if we're not sharing */
Paul Mundtdd8632a2009-01-08 12:04:47 +00001139 BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1140 vma->vm_region->vm_top = vma->vm_region->vm_end;
David Howells645d83c2009-09-24 15:13:10 +01001141 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142 }
Paul Mundtdd8632a2009-01-08 12:04:47 +00001143 if (ret != -ENOSYS)
1144 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145
1146 /* getting an ENOSYS error indicates that direct mmap isn't
1147 * possible (as opposed to tried but failed) so we'll try to
1148 * make a private copy of the data and map that instead */
1149 }
1150
David Howells8feae132009-01-08 12:04:47 +00001151 rlen = PAGE_ALIGN(len);
1152
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 */
David Howells8feae132009-01-08 12:04:47 +00001157 order = get_order(rlen);
1158 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
David Howells8feae132009-01-08 12:04:47 +00001167 point = rlen >> 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;
1189 region->vm_end = region->vm_start + rlen;
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);
David Howells8feae132009-01-08 12:04:47 +00001205 ret = vma->vm_file->f_op->read(vma->vm_file, base, rlen, &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 */
David Howells8feae132009-01-08 12:04:47 +00001212 if (ret < rlen)
1213 memset(base + ret, 0, rlen - ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 }
1216
1217 return 0;
1218
1219error_free:
David Howells8feae132009-01-08 12:04:47 +00001220 free_page_series(region->vm_start, region->vm_end);
1221 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;
1262
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 /* we've determined that we can make the mapping, now translate what we
1264 * now know into VMA flags */
1265 vm_flags = determine_vm_flags(file, prot, flags, capabilities);
1266
David Howells8feae132009-01-08 12:04:47 +00001267 /* we're going to need to record the mapping */
1268 region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1269 if (!region)
1270 goto error_getting_region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271
David Howells8feae132009-01-08 12:04:47 +00001272 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1273 if (!vma)
1274 goto error_getting_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275
David Howells1e2ae592010-01-15 17:01:33 -08001276 region->vm_usage = 1;
David Howells8feae132009-01-08 12:04:47 +00001277 region->vm_flags = vm_flags;
1278 region->vm_pgoff = pgoff;
1279
Rik van Riel5beb4932010-03-05 13:42:07 -08001280 INIT_LIST_HEAD(&vma->anon_vma_chain);
David Howells8feae132009-01-08 12:04:47 +00001281 vma->vm_flags = vm_flags;
1282 vma->vm_pgoff = pgoff;
1283
1284 if (file) {
1285 region->vm_file = file;
1286 get_file(file);
1287 vma->vm_file = file;
1288 get_file(file);
1289 if (vm_flags & VM_EXECUTABLE) {
1290 added_exe_file_vma(current->mm);
1291 vma->vm_mm = current->mm;
1292 }
1293 }
1294
1295 down_write(&nommu_region_sem);
1296
1297 /* if we want to share, we need to check for regions created by other
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 * mmap() calls that overlap with our proposed mapping
David Howells8feae132009-01-08 12:04:47 +00001299 * - we can only share with a superset match on most regular files
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300 * - shared mappings on character devices and memory backed files are
1301 * permitted to overlap inexactly as far as we are concerned for in
1302 * these cases, sharing is handled in the driver or filesystem rather
1303 * than here
1304 */
1305 if (vm_flags & VM_MAYSHARE) {
David Howells8feae132009-01-08 12:04:47 +00001306 struct vm_region *pregion;
1307 unsigned long pglen, rpglen, pgend, rpgend, start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308
David Howells8feae132009-01-08 12:04:47 +00001309 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1310 pgend = pgoff + pglen;
David Howells165b2392007-03-22 00:11:24 -08001311
David Howells8feae132009-01-08 12:04:47 +00001312 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1313 pregion = rb_entry(rb, struct vm_region, vm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314
David Howells8feae132009-01-08 12:04:47 +00001315 if (!(pregion->vm_flags & VM_MAYSHARE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 continue;
1317
1318 /* search for overlapping mappings on the same file */
David Howells8feae132009-01-08 12:04:47 +00001319 if (pregion->vm_file->f_path.dentry->d_inode !=
1320 file->f_path.dentry->d_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321 continue;
1322
David Howells8feae132009-01-08 12:04:47 +00001323 if (pregion->vm_pgoff >= pgend)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324 continue;
1325
David Howells8feae132009-01-08 12:04:47 +00001326 rpglen = pregion->vm_end - pregion->vm_start;
1327 rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1328 rpgend = pregion->vm_pgoff + rpglen;
1329 if (pgoff >= rpgend)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330 continue;
1331
David Howells8feae132009-01-08 12:04:47 +00001332 /* handle inexactly overlapping matches between
1333 * mappings */
1334 if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1335 !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1336 /* new mapping is not a subset of the region */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 if (!(capabilities & BDI_CAP_MAP_DIRECT))
1338 goto sharing_violation;
1339 continue;
1340 }
1341
David Howells8feae132009-01-08 12:04:47 +00001342 /* we've found a region we can share */
David Howells1e2ae592010-01-15 17:01:33 -08001343 pregion->vm_usage++;
David Howells8feae132009-01-08 12:04:47 +00001344 vma->vm_region = pregion;
1345 start = pregion->vm_start;
1346 start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1347 vma->vm_start = start;
1348 vma->vm_end = start + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349
David Howells8feae132009-01-08 12:04:47 +00001350 if (pregion->vm_flags & VM_MAPPED_COPY) {
1351 kdebug("share copy");
1352 vma->vm_flags |= VM_MAPPED_COPY;
1353 } else {
1354 kdebug("share mmap");
1355 ret = do_mmap_shared_file(vma);
1356 if (ret < 0) {
1357 vma->vm_region = NULL;
1358 vma->vm_start = 0;
1359 vma->vm_end = 0;
David Howells1e2ae592010-01-15 17:01:33 -08001360 pregion->vm_usage--;
David Howells8feae132009-01-08 12:04:47 +00001361 pregion = NULL;
1362 goto error_just_free;
1363 }
1364 }
1365 fput(region->vm_file);
1366 kmem_cache_free(vm_region_jar, region);
1367 region = pregion;
1368 result = start;
1369 goto share;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 }
1371
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 /* obtain the address at which to make a shared mapping
1373 * - this is the hook for quasi-memory character devices to
1374 * tell us the location of a shared mapping
1375 */
David Howells645d83c2009-09-24 15:13:10 +01001376 if (capabilities & BDI_CAP_MAP_DIRECT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 addr = file->f_op->get_unmapped_area(file, addr, len,
1378 pgoff, flags);
1379 if (IS_ERR((void *) addr)) {
1380 ret = addr;
1381 if (ret != (unsigned long) -ENOSYS)
David Howells8feae132009-01-08 12:04:47 +00001382 goto error_just_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383
1384 /* the driver refused to tell us where to site
1385 * the mapping so we'll have to attempt to copy
1386 * it */
1387 ret = (unsigned long) -ENODEV;
1388 if (!(capabilities & BDI_CAP_MAP_COPY))
David Howells8feae132009-01-08 12:04:47 +00001389 goto error_just_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390
1391 capabilities &= ~BDI_CAP_MAP_DIRECT;
David Howells8feae132009-01-08 12:04:47 +00001392 } else {
1393 vma->vm_start = region->vm_start = addr;
1394 vma->vm_end = region->vm_end = addr + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 }
1396 }
1397 }
1398
David Howells8feae132009-01-08 12:04:47 +00001399 vma->vm_region = region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400
David Howells645d83c2009-09-24 15:13:10 +01001401 /* set up the mapping
1402 * - the region is filled in if BDI_CAP_MAP_DIRECT is still set
1403 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 if (file && vma->vm_flags & VM_SHARED)
David Howells8feae132009-01-08 12:04:47 +00001405 ret = do_mmap_shared_file(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 else
David Howells645d83c2009-09-24 15:13:10 +01001407 ret = do_mmap_private(vma, region, len, capabilities);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 if (ret < 0)
David Howells645d83c2009-09-24 15:13:10 +01001409 goto error_just_free;
1410 add_nommu_region(region);
David Howells8feae132009-01-08 12:04:47 +00001411
Jie Zhangea637632009-12-14 18:00:02 -08001412 /* clear anonymous mappings that don't ask for uninitialized data */
1413 if (!vma->vm_file && !(flags & MAP_UNINITIALIZED))
1414 memset((void *)region->vm_start, 0,
1415 region->vm_end - region->vm_start);
1416
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 /* okay... we have a mapping; now we have to register it */
David Howells8feae132009-01-08 12:04:47 +00001418 result = vma->vm_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420 current->mm->total_vm += len >> PAGE_SHIFT;
1421
David Howells8feae132009-01-08 12:04:47 +00001422share:
1423 add_vma_to_mm(current->mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424
Mike Frysingercfe79c02010-01-06 17:23:23 +00001425 /* we flush the region from the icache only when the first executable
1426 * mapping of it is made */
1427 if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1428 flush_icache_range(region->vm_start, region->vm_end);
1429 region->vm_icache_flushed = true;
1430 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431
Mike Frysingercfe79c02010-01-06 17:23:23 +00001432 up_write(&nommu_region_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433
David Howells8feae132009-01-08 12:04:47 +00001434 kleave(" = %lx", result);
1435 return result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436
David Howells8feae132009-01-08 12:04:47 +00001437error_just_free:
1438 up_write(&nommu_region_sem);
1439error:
David Howells89a86402009-10-30 13:13:26 +00001440 if (region->vm_file)
1441 fput(region->vm_file);
David Howells8feae132009-01-08 12:04:47 +00001442 kmem_cache_free(vm_region_jar, region);
David Howells89a86402009-10-30 13:13:26 +00001443 if (vma->vm_file)
1444 fput(vma->vm_file);
David Howells8feae132009-01-08 12:04:47 +00001445 if (vma->vm_flags & VM_EXECUTABLE)
1446 removed_exe_file_vma(vma->vm_mm);
1447 kmem_cache_free(vm_area_cachep, vma);
1448 kleave(" = %d", ret);
1449 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450
David Howells8feae132009-01-08 12:04:47 +00001451sharing_violation:
1452 up_write(&nommu_region_sem);
1453 printk(KERN_WARNING "Attempt to share mismatched mappings\n");
1454 ret = -EINVAL;
1455 goto error;
1456
1457error_getting_vma:
1458 kmem_cache_free(vm_region_jar, region);
1459 printk(KERN_WARNING "Allocation of vma for %lu byte allocation"
1460 " from process %d failed\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 len, current->pid);
David Rientjes7bf02ea2011-05-24 17:11:16 -07001462 show_free_areas(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463 return -ENOMEM;
1464
David Howells8feae132009-01-08 12:04:47 +00001465error_getting_region:
1466 printk(KERN_WARNING "Allocation of vm region for %lu byte allocation"
1467 " from process %d failed\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 len, current->pid);
David Rientjes7bf02ea2011-05-24 17:11:16 -07001469 show_free_areas(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 return -ENOMEM;
1471}
Paul Mundtb5073172007-07-21 04:37:25 -07001472EXPORT_SYMBOL(do_mmap_pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001474SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1475 unsigned long, prot, unsigned long, flags,
1476 unsigned long, fd, unsigned long, pgoff)
1477{
1478 struct file *file = NULL;
1479 unsigned long retval = -EBADF;
1480
Al Viro120a7952010-10-30 02:54:44 -04001481 audit_mmap_fd(fd, flags);
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001482 if (!(flags & MAP_ANONYMOUS)) {
1483 file = fget(fd);
1484 if (!file)
1485 goto out;
1486 }
1487
1488 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1489
1490 down_write(&current->mm->mmap_sem);
1491 retval = do_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1492 up_write(&current->mm->mmap_sem);
1493
1494 if (file)
1495 fput(file);
1496out:
1497 return retval;
1498}
1499
Christoph Hellwiga4679372010-03-10 15:21:15 -08001500#ifdef __ARCH_WANT_SYS_OLD_MMAP
1501struct mmap_arg_struct {
1502 unsigned long addr;
1503 unsigned long len;
1504 unsigned long prot;
1505 unsigned long flags;
1506 unsigned long fd;
1507 unsigned long offset;
1508};
1509
1510SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1511{
1512 struct mmap_arg_struct a;
1513
1514 if (copy_from_user(&a, arg, sizeof(a)))
1515 return -EFAULT;
1516 if (a.offset & ~PAGE_MASK)
1517 return -EINVAL;
1518
1519 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1520 a.offset >> PAGE_SHIFT);
1521}
1522#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1523
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524/*
David Howells8feae132009-01-08 12:04:47 +00001525 * split a vma into two pieces at address 'addr', a new vma is allocated either
1526 * for the first part or the tail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 */
David Howells8feae132009-01-08 12:04:47 +00001528int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1529 unsigned long addr, int new_below)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530{
David Howells8feae132009-01-08 12:04:47 +00001531 struct vm_area_struct *new;
1532 struct vm_region *region;
1533 unsigned long npages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534
David Howells8feae132009-01-08 12:04:47 +00001535 kenter("");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536
David Howells779c1022010-01-15 17:01:34 -08001537 /* we're only permitted to split anonymous regions (these should have
1538 * only a single usage on the region) */
1539 if (vma->vm_file)
David Howells8feae132009-01-08 12:04:47 +00001540 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541
David Howells8feae132009-01-08 12:04:47 +00001542 if (mm->map_count >= sysctl_max_map_count)
1543 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544
David Howells8feae132009-01-08 12:04:47 +00001545 region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1546 if (!region)
1547 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548
David Howells8feae132009-01-08 12:04:47 +00001549 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1550 if (!new) {
1551 kmem_cache_free(vm_region_jar, region);
1552 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553 }
David Howells8feae132009-01-08 12:04:47 +00001554
1555 /* most fields are the same, copy all, and then fixup */
1556 *new = *vma;
1557 *region = *vma->vm_region;
1558 new->vm_region = region;
1559
1560 npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1561
1562 if (new_below) {
Paul Mundtdd8632a2009-01-08 12:04:47 +00001563 region->vm_top = region->vm_end = new->vm_end = addr;
David Howells8feae132009-01-08 12:04:47 +00001564 } else {
1565 region->vm_start = new->vm_start = addr;
1566 region->vm_pgoff = new->vm_pgoff += npages;
1567 }
1568
1569 if (new->vm_ops && new->vm_ops->open)
1570 new->vm_ops->open(new);
1571
1572 delete_vma_from_mm(vma);
1573 down_write(&nommu_region_sem);
1574 delete_nommu_region(vma->vm_region);
1575 if (new_below) {
1576 vma->vm_region->vm_start = vma->vm_start = addr;
1577 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1578 } else {
1579 vma->vm_region->vm_end = vma->vm_end = addr;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001580 vma->vm_region->vm_top = addr;
David Howells8feae132009-01-08 12:04:47 +00001581 }
1582 add_nommu_region(vma->vm_region);
1583 add_nommu_region(new->vm_region);
1584 up_write(&nommu_region_sem);
1585 add_vma_to_mm(mm, vma);
1586 add_vma_to_mm(mm, new);
1587 return 0;
1588}
1589
1590/*
1591 * shrink a VMA by removing the specified chunk from either the beginning or
1592 * the end
1593 */
1594static int shrink_vma(struct mm_struct *mm,
1595 struct vm_area_struct *vma,
1596 unsigned long from, unsigned long to)
1597{
1598 struct vm_region *region;
1599
1600 kenter("");
1601
1602 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1603 * and list */
1604 delete_vma_from_mm(vma);
1605 if (from > vma->vm_start)
1606 vma->vm_end = from;
1607 else
1608 vma->vm_start = to;
1609 add_vma_to_mm(mm, vma);
1610
1611 /* cut the backing region down to size */
1612 region = vma->vm_region;
David Howells1e2ae592010-01-15 17:01:33 -08001613 BUG_ON(region->vm_usage != 1);
David Howells8feae132009-01-08 12:04:47 +00001614
1615 down_write(&nommu_region_sem);
1616 delete_nommu_region(region);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001617 if (from > region->vm_start) {
1618 to = region->vm_top;
1619 region->vm_top = region->vm_end = from;
1620 } else {
David Howells8feae132009-01-08 12:04:47 +00001621 region->vm_start = to;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001622 }
David Howells8feae132009-01-08 12:04:47 +00001623 add_nommu_region(region);
1624 up_write(&nommu_region_sem);
1625
1626 free_page_series(from, to);
1627 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628}
1629
David Howells30340972006-09-27 01:50:20 -07001630/*
1631 * release a mapping
David Howells8feae132009-01-08 12:04:47 +00001632 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1633 * VMA, though it need not cover the whole VMA
David Howells30340972006-09-27 01:50:20 -07001634 */
David Howells8feae132009-01-08 12:04:47 +00001635int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636{
David Howells8feae132009-01-08 12:04:47 +00001637 struct vm_area_struct *vma;
David Howells8feae132009-01-08 12:04:47 +00001638 unsigned long end = start + len;
1639 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640
David Howells8feae132009-01-08 12:04:47 +00001641 kenter(",%lx,%zx", start, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642
David Howells8feae132009-01-08 12:04:47 +00001643 if (len == 0)
1644 return -EINVAL;
1645
1646 /* find the first potentially overlapping VMA */
1647 vma = find_vma(mm, start);
1648 if (!vma) {
David Howells33e5d7692009-04-02 16:56:32 -07001649 static int limit = 0;
1650 if (limit < 5) {
1651 printk(KERN_WARNING
1652 "munmap of memory not mmapped by process %d"
1653 " (%s): 0x%lx-0x%lx\n",
1654 current->pid, current->comm,
1655 start, start + len - 1);
1656 limit++;
1657 }
David Howells8feae132009-01-08 12:04:47 +00001658 return -EINVAL;
David Howells30340972006-09-27 01:50:20 -07001659 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660
David Howells8feae132009-01-08 12:04:47 +00001661 /* we're allowed to split an anonymous VMA but not a file-backed one */
1662 if (vma->vm_file) {
1663 do {
1664 if (start > vma->vm_start) {
1665 kleave(" = -EINVAL [miss]");
1666 return -EINVAL;
1667 }
1668 if (end == vma->vm_end)
1669 goto erase_whole_vma;
Namhyung Kimd75a3102011-05-24 17:11:25 -07001670 vma = vma->vm_next;
1671 } while (vma);
David Howells8feae132009-01-08 12:04:47 +00001672 kleave(" = -EINVAL [split file]");
1673 return -EINVAL;
1674 } else {
1675 /* the chunk must be a subset of the VMA found */
1676 if (start == vma->vm_start && end == vma->vm_end)
1677 goto erase_whole_vma;
1678 if (start < vma->vm_start || end > vma->vm_end) {
1679 kleave(" = -EINVAL [superset]");
1680 return -EINVAL;
1681 }
1682 if (start & ~PAGE_MASK) {
1683 kleave(" = -EINVAL [unaligned start]");
1684 return -EINVAL;
1685 }
1686 if (end != vma->vm_end && end & ~PAGE_MASK) {
1687 kleave(" = -EINVAL [unaligned split]");
1688 return -EINVAL;
1689 }
1690 if (start != vma->vm_start && end != vma->vm_end) {
1691 ret = split_vma(mm, vma, start, 1);
1692 if (ret < 0) {
1693 kleave(" = %d [split]", ret);
1694 return ret;
1695 }
1696 }
1697 return shrink_vma(mm, vma, start, end);
1698 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699
David Howells8feae132009-01-08 12:04:47 +00001700erase_whole_vma:
1701 delete_vma_from_mm(vma);
1702 delete_vma(mm, vma);
1703 kleave(" = 0");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704 return 0;
1705}
Paul Mundtb5073172007-07-21 04:37:25 -07001706EXPORT_SYMBOL(do_munmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707
Heiko Carstens6a6160a2009-01-14 14:14:15 +01001708SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
David Howells30340972006-09-27 01:50:20 -07001709{
1710 int ret;
1711 struct mm_struct *mm = current->mm;
1712
1713 down_write(&mm->mmap_sem);
1714 ret = do_munmap(mm, addr, len);
1715 up_write(&mm->mmap_sem);
1716 return ret;
1717}
1718
1719/*
David Howells8feae132009-01-08 12:04:47 +00001720 * release all the mappings made in a process's VM space
David Howells30340972006-09-27 01:50:20 -07001721 */
David Howells8feae132009-01-08 12:04:47 +00001722void exit_mmap(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723{
David Howells8feae132009-01-08 12:04:47 +00001724 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725
David Howells8feae132009-01-08 12:04:47 +00001726 if (!mm)
1727 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728
David Howells8feae132009-01-08 12:04:47 +00001729 kenter("");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730
David Howells8feae132009-01-08 12:04:47 +00001731 mm->total_vm = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732
David Howells8feae132009-01-08 12:04:47 +00001733 while ((vma = mm->mmap)) {
1734 mm->mmap = vma->vm_next;
1735 delete_vma_from_mm(vma);
1736 delete_vma(mm, vma);
Steven J. Magnani04c34962010-11-24 12:56:54 -08001737 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738 }
David Howells8feae132009-01-08 12:04:47 +00001739
1740 kleave("");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741}
1742
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743unsigned long do_brk(unsigned long addr, unsigned long len)
1744{
1745 return -ENOMEM;
1746}
1747
1748/*
David Howells6fa5f802006-09-27 01:50:21 -07001749 * expand (or shrink) an existing mapping, potentially moving it at the same
1750 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751 *
David Howells6fa5f802006-09-27 01:50:21 -07001752 * under NOMMU conditions, we only permit changing a mapping's size, and only
David Howells8feae132009-01-08 12:04:47 +00001753 * as long as it stays within the region allocated by do_mmap_private() and the
1754 * block is not shareable
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 *
David Howells6fa5f802006-09-27 01:50:21 -07001756 * MREMAP_FIXED is not supported under NOMMU conditions
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757 */
1758unsigned long do_mremap(unsigned long addr,
1759 unsigned long old_len, unsigned long new_len,
1760 unsigned long flags, unsigned long new_addr)
1761{
David Howells6fa5f802006-09-27 01:50:21 -07001762 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763
1764 /* insanity checks first */
David Howells8feae132009-01-08 12:04:47 +00001765 if (old_len == 0 || new_len == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 return (unsigned long) -EINVAL;
1767
David Howells8feae132009-01-08 12:04:47 +00001768 if (addr & ~PAGE_MASK)
1769 return -EINVAL;
1770
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771 if (flags & MREMAP_FIXED && new_addr != addr)
1772 return (unsigned long) -EINVAL;
1773
David Howells8feae132009-01-08 12:04:47 +00001774 vma = find_vma_exact(current->mm, addr, old_len);
David Howells6fa5f802006-09-27 01:50:21 -07001775 if (!vma)
1776 return (unsigned long) -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777
David Howells6fa5f802006-09-27 01:50:21 -07001778 if (vma->vm_end != vma->vm_start + old_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779 return (unsigned long) -EFAULT;
1780
David Howells6fa5f802006-09-27 01:50:21 -07001781 if (vma->vm_flags & VM_MAYSHARE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782 return (unsigned long) -EPERM;
1783
David Howells8feae132009-01-08 12:04:47 +00001784 if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785 return (unsigned long) -ENOMEM;
1786
1787 /* all checks complete - do it */
David Howells6fa5f802006-09-27 01:50:21 -07001788 vma->vm_end = vma->vm_start + new_len;
David Howells6fa5f802006-09-27 01:50:21 -07001789 return vma->vm_start;
1790}
Paul Mundtb5073172007-07-21 04:37:25 -07001791EXPORT_SYMBOL(do_mremap);
David Howells6fa5f802006-09-27 01:50:21 -07001792
Heiko Carstens6a6160a2009-01-14 14:14:15 +01001793SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1794 unsigned long, new_len, unsigned long, flags,
1795 unsigned long, new_addr)
David Howells6fa5f802006-09-27 01:50:21 -07001796{
1797 unsigned long ret;
1798
1799 down_write(&current->mm->mmap_sem);
1800 ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1801 up_write(&current->mm->mmap_sem);
1802 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803}
1804
Linus Torvalds6aab3412005-11-28 14:34:23 -08001805struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001806 unsigned int foll_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807{
1808 return NULL;
1809}
1810
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811int remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
1812 unsigned long to, unsigned long size, pgprot_t prot)
1813{
Greg Ungerer66aa2b42005-09-12 11:18:10 +10001814 vma->vm_start = vma->vm_pgoff << PAGE_SHIFT;
1815 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816}
Luke Yang22c4af42006-07-14 00:24:09 -07001817EXPORT_SYMBOL(remap_pfn_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818
Paul Mundtf905bc42008-02-04 22:29:59 -08001819int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1820 unsigned long pgoff)
1821{
1822 unsigned int size = vma->vm_end - vma->vm_start;
1823
1824 if (!(vma->vm_flags & VM_USERMAP))
1825 return -EINVAL;
1826
1827 vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1828 vma->vm_end = vma->vm_start + size;
1829
1830 return 0;
1831}
1832EXPORT_SYMBOL(remap_vmalloc_range);
1833
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1835 unsigned long len, unsigned long pgoff, unsigned long flags)
1836{
1837 return -ENOMEM;
1838}
1839
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001840void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841{
1842}
1843
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844void unmap_mapping_range(struct address_space *mapping,
1845 loff_t const holebegin, loff_t const holelen,
1846 int even_cows)
1847{
1848}
Luke Yang22c4af42006-07-14 00:24:09 -07001849EXPORT_SYMBOL(unmap_mapping_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850
1851/*
1852 * Check that a process has enough memory to allocate a new virtual
1853 * mapping. 0 means there is enough memory for the allocation to
1854 * succeed and -ENOMEM implies there is not.
1855 *
1856 * We currently support three overcommit policies, which are set via the
1857 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
1858 *
1859 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1860 * Additional code 2002 Jul 20 by Robert Love.
1861 *
1862 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1863 *
1864 * Note this is a helper function intended to be used by LSMs which
1865 * wish to use this logic.
1866 */
Alan Cox34b4e4a2007-08-22 14:01:28 -07001867int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868{
1869 unsigned long free, allowed;
1870
1871 vm_acct_memory(pages);
1872
1873 /*
1874 * Sometimes we want to use more memory than we have
1875 */
1876 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1877 return 0;
1878
1879 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1880 unsigned long n;
1881
Christoph Lameter347ce432006-06-30 01:55:35 -07001882 free = global_page_state(NR_FILE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883 free += nr_swap_pages;
1884
1885 /*
1886 * Any slabs which are created with the
1887 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1888 * which are reclaimable, under pressure. The dentry
1889 * cache and most inode caches should fall into this
1890 */
Christoph Lameter972d1a72006-09-25 23:31:51 -07001891 free += global_page_state(NR_SLAB_RECLAIMABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892
1893 /*
1894 * Leave the last 3% for root
1895 */
1896 if (!cap_sys_admin)
1897 free -= free / 32;
1898
1899 if (free > pages)
1900 return 0;
1901
1902 /*
1903 * nr_free_pages() is very expensive on large systems,
1904 * only call if we're about to fail.
1905 */
1906 n = nr_free_pages();
Hideo AOKId5ddc792006-04-10 22:53:01 -07001907
1908 /*
1909 * Leave reserved pages. The pages are not for anonymous pages.
1910 */
1911 if (n <= totalreserve_pages)
1912 goto error;
1913 else
1914 n -= totalreserve_pages;
1915
1916 /*
1917 * Leave the last 3% for root
1918 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919 if (!cap_sys_admin)
1920 n -= n / 32;
1921 free += n;
1922
1923 if (free > pages)
1924 return 0;
Hideo AOKId5ddc792006-04-10 22:53:01 -07001925
1926 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927 }
1928
1929 allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1930 /*
1931 * Leave the last 3% for root
1932 */
1933 if (!cap_sys_admin)
1934 allowed -= allowed / 32;
1935 allowed += total_swap_pages;
1936
1937 /* Don't let a single process grow too big:
1938 leave 3% of the size of this process for other processes */
Alan Cox731572d2008-10-29 14:01:20 -07001939 if (mm)
1940 allowed -= mm->total_vm / 32;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07001942 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943 return 0;
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07001944
Hideo AOKId5ddc792006-04-10 22:53:01 -07001945error:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 vm_unacct_memory(pages);
1947
1948 return -ENOMEM;
1949}
1950
Stephen Wilsoncae5d392011-03-13 15:49:17 -04001951int in_gate_area_no_mm(unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952{
1953 return 0;
1954}
David Howellsb0e15192006-01-06 00:11:42 -08001955
Nick Piggind0217ac2007-07-19 01:47:03 -07001956int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
David Howellsb0e15192006-01-06 00:11:42 -08001957{
1958 BUG();
Nick Piggind0217ac2007-07-19 01:47:03 -07001959 return 0;
David Howellsb0e15192006-01-06 00:11:42 -08001960}
Paul Mundtb5073172007-07-21 04:37:25 -07001961EXPORT_SYMBOL(filemap_fault);
David Howells0ec76a12006-09-27 01:50:15 -07001962
Mike Frysingerf55f1992011-03-29 14:05:12 +01001963static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
1964 unsigned long addr, void *buf, int len, int write)
David Howells0ec76a12006-09-27 01:50:15 -07001965{
David Howells0ec76a12006-09-27 01:50:15 -07001966 struct vm_area_struct *vma;
David Howells0ec76a12006-09-27 01:50:15 -07001967
1968 down_read(&mm->mmap_sem);
1969
1970 /* the access must start within one of the target process's mappings */
David Howells0159b142006-09-27 01:50:16 -07001971 vma = find_vma(mm, addr);
1972 if (vma) {
David Howells0ec76a12006-09-27 01:50:15 -07001973 /* don't overrun this mapping */
1974 if (addr + len >= vma->vm_end)
1975 len = vma->vm_end - addr;
1976
1977 /* only read or write mappings where it is permitted */
David Howellsd00c7b992006-09-27 01:50:19 -07001978 if (write && vma->vm_flags & VM_MAYWRITE)
Jie Zhang79597222010-01-06 17:23:28 +00001979 copy_to_user_page(vma, NULL, addr,
1980 (void *) addr, buf, len);
David Howellsd00c7b992006-09-27 01:50:19 -07001981 else if (!write && vma->vm_flags & VM_MAYREAD)
Jie Zhang79597222010-01-06 17:23:28 +00001982 copy_from_user_page(vma, NULL, addr,
1983 buf, (void *) addr, len);
David Howells0ec76a12006-09-27 01:50:15 -07001984 else
1985 len = 0;
1986 } else {
1987 len = 0;
1988 }
1989
1990 up_read(&mm->mmap_sem);
Mike Frysingerf55f1992011-03-29 14:05:12 +01001991
1992 return len;
1993}
1994
1995/**
1996 * @access_remote_vm - access another process' address space
1997 * @mm: the mm_struct of the target address space
1998 * @addr: start address to access
1999 * @buf: source or destination buffer
2000 * @len: number of bytes to transfer
2001 * @write: whether the access is a write
2002 *
2003 * The caller must hold a reference on @mm.
2004 */
2005int access_remote_vm(struct mm_struct *mm, unsigned long addr,
2006 void *buf, int len, int write)
2007{
2008 return __access_remote_vm(NULL, mm, addr, buf, len, write);
2009}
2010
2011/*
2012 * Access another process' address space.
2013 * - source/target buffer must be kernel space
2014 */
2015int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
2016{
2017 struct mm_struct *mm;
2018
2019 if (addr + len < addr)
2020 return 0;
2021
2022 mm = get_task_mm(tsk);
2023 if (!mm)
2024 return 0;
2025
2026 len = __access_remote_vm(tsk, mm, addr, buf, len, write);
2027
David Howells0ec76a12006-09-27 01:50:15 -07002028 mmput(mm);
2029 return len;
2030}
David Howells7e660872010-01-15 17:01:39 -08002031
2032/**
2033 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
2034 * @inode: The inode to check
2035 * @size: The current filesize of the inode
2036 * @newsize: The proposed filesize of the inode
2037 *
2038 * Check the shared mappings on an inode on behalf of a shrinking truncate to
2039 * make sure that that any outstanding VMAs aren't broken and then shrink the
2040 * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
2041 * automatically grant mappings that are too large.
2042 */
2043int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
2044 size_t newsize)
2045{
2046 struct vm_area_struct *vma;
2047 struct prio_tree_iter iter;
2048 struct vm_region *region;
2049 pgoff_t low, high;
2050 size_t r_size, r_top;
2051
2052 low = newsize >> PAGE_SHIFT;
2053 high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
2054
2055 down_write(&nommu_region_sem);
2056
2057 /* search for VMAs that fall within the dead zone */
2058 vma_prio_tree_foreach(vma, &iter, &inode->i_mapping->i_mmap,
2059 low, high) {
2060 /* found one - only interested if it's shared out of the page
2061 * cache */
2062 if (vma->vm_flags & VM_SHARED) {
2063 up_write(&nommu_region_sem);
2064 return -ETXTBSY; /* not quite true, but near enough */
2065 }
2066 }
2067
2068 /* reduce any regions that overlap the dead zone - if in existence,
2069 * these will be pointed to by VMAs that don't overlap the dead zone
2070 *
2071 * we don't check for any regions that start beyond the EOF as there
2072 * shouldn't be any
2073 */
2074 vma_prio_tree_foreach(vma, &iter, &inode->i_mapping->i_mmap,
2075 0, ULONG_MAX) {
2076 if (!(vma->vm_flags & VM_SHARED))
2077 continue;
2078
2079 region = vma->vm_region;
2080 r_size = region->vm_top - region->vm_start;
2081 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
2082
2083 if (r_top > newsize) {
2084 region->vm_top -= r_top - newsize;
2085 if (region->vm_end > region->vm_top)
2086 region->vm_end = region->vm_top;
2087 }
2088 }
2089
2090 up_write(&nommu_region_sem);
2091 return 0;
2092}