blob: 605ace8982a8dcbc14ac061ab824166a16220180 [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 Mundteb6434d2009-01-21 17:45:47 +090013 * Copyright (c) 2007-2009 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>
32
33#include <asm/uaccess.h>
34#include <asm/tlb.h>
35#include <asm/tlbflush.h>
Bernd Schmidteb8cdec2009-09-21 17:03:57 -070036#include <asm/mmu_context.h>
David Howells8feae132009-01-08 12:04:47 +000037#include "internal.h"
38
39static inline __attribute__((format(printf, 1, 2)))
40void no_printk(const char *fmt, ...)
41{
42}
43
44#if 0
45#define kenter(FMT, ...) \
46 printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
47#define kleave(FMT, ...) \
48 printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
49#define kdebug(FMT, ...) \
50 printk(KERN_DEBUG "xxx" FMT"yyy\n", ##__VA_ARGS__)
51#else
52#define kenter(FMT, ...) \
53 no_printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
54#define kleave(FMT, ...) \
55 no_printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
56#define kdebug(FMT, ...) \
57 no_printk(KERN_DEBUG FMT"\n", ##__VA_ARGS__)
58#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -070059
60void *high_memory;
61struct page *mem_map;
62unsigned long max_mapnr;
63unsigned long num_physpages;
Hugh Dickins4266c972009-09-23 17:05:53 +010064unsigned long highest_memmap_pfn;
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -070065struct percpu_counter vm_committed_as;
Linus Torvalds1da177e2005-04-16 15:20:36 -070066int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
67int sysctl_overcommit_ratio = 50; /* default is 50% */
68int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
David Howellsfc4d5c22009-05-06 16:03:05 -070069int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -070070int heap_stack_gap = 0;
71
David Howells33e5d7692009-04-02 16:56:32 -070072atomic_long_t mmap_pages_allocated;
David Howells8feae132009-01-08 12:04:47 +000073
Linus Torvalds1da177e2005-04-16 15:20:36 -070074EXPORT_SYMBOL(mem_map);
Wu, Bryan6a04de62007-04-11 23:28:47 -070075EXPORT_SYMBOL(num_physpages);
Linus Torvalds1da177e2005-04-16 15:20:36 -070076
David Howells8feae132009-01-08 12:04:47 +000077/* list of mapped, potentially shareable regions */
78static struct kmem_cache *vm_region_jar;
79struct rb_root nommu_region_tree = RB_ROOT;
80DECLARE_RWSEM(nommu_region_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -070081
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +040082const struct vm_operations_struct generic_file_vm_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -070083};
84
85/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070086 * Return the total memory allocated for this pointer, not
87 * just what the caller asked for.
88 *
89 * Doesn't have to be accurate, i.e. may have races.
90 */
91unsigned int kobjsize(const void *objp)
92{
93 struct page *page;
94
Michael Hennerich4016a132008-04-28 02:13:38 -070095 /*
96 * If the object we have should not have ksize performed on it,
97 * return size of 0
98 */
Paul Mundt5a1603b2008-06-12 16:29:55 +090099 if (!objp || !virt_addr_valid(objp))
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700100 return 0;
101
102 page = virt_to_head_page(objp);
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700103
104 /*
105 * If the allocator sets PageSlab, we know the pointer came from
106 * kmalloc().
107 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 if (PageSlab(page))
109 return ksize(objp);
110
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700111 /*
Paul Mundtab2e83e2009-01-08 12:04:48 +0000112 * If it's not a compound page, see if we have a matching VMA
113 * region. This test is intentionally done in reverse order,
114 * so if there's no VMA, we still fall through and hand back
115 * PAGE_SIZE for 0-order pages.
116 */
117 if (!PageCompound(page)) {
118 struct vm_area_struct *vma;
119
120 vma = find_vma(current->mm, (unsigned long)objp);
121 if (vma)
122 return vma->vm_end - vma->vm_start;
123 }
124
125 /*
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700126 * The ksize() function is only guaranteed to work for pointers
Paul Mundt5a1603b2008-06-12 16:29:55 +0900127 * returned by kmalloc(). So handle arbitrary pointers here.
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700128 */
Paul Mundt5a1603b2008-06-12 16:29:55 +0900129 return PAGE_SIZE << compound_order(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130}
131
Nick Pigginb291f002008-10-18 20:26:44 -0700132int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
Hugh Dickins4266c972009-09-23 17:05:53 +0100133 unsigned long start, int nr_pages, unsigned int foll_flags,
Peter Zijlstra9d737772009-06-25 11:58:55 +0200134 struct page **pages, struct vm_area_struct **vmas)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135{
Sonic Zhang910e46d2006-09-27 01:50:17 -0700136 struct vm_area_struct *vma;
David Howells7b4d5b82006-09-27 01:50:18 -0700137 unsigned long vm_flags;
138 int i;
139
140 /* calculate required read or write permissions.
Hugh Dickins58fa8792009-09-21 17:03:31 -0700141 * If FOLL_FORCE is set, we only require the "MAY" flags.
David Howells7b4d5b82006-09-27 01:50:18 -0700142 */
Hugh Dickins58fa8792009-09-21 17:03:31 -0700143 vm_flags = (foll_flags & FOLL_WRITE) ?
144 (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
145 vm_flags &= (foll_flags & FOLL_FORCE) ?
146 (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147
Peter Zijlstra9d737772009-06-25 11:58:55 +0200148 for (i = 0; i < nr_pages; i++) {
Steven J. Magnanic08c6e12010-03-05 13:42:24 -0800149 vma = find_extend_vma(mm, start);
David Howells7b4d5b82006-09-27 01:50:18 -0700150 if (!vma)
151 goto finish_or_fault;
152
153 /* protect what we can, including chardevs */
Hugh Dickins1c3aff12009-09-21 17:03:24 -0700154 if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
155 !(vm_flags & vma->vm_flags))
David Howells7b4d5b82006-09-27 01:50:18 -0700156 goto finish_or_fault;
Sonic Zhang910e46d2006-09-27 01:50:17 -0700157
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 if (pages) {
159 pages[i] = virt_to_page(start);
160 if (pages[i])
161 page_cache_get(pages[i]);
162 }
163 if (vmas)
Sonic Zhang910e46d2006-09-27 01:50:17 -0700164 vmas[i] = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165 start += PAGE_SIZE;
166 }
David Howells7b4d5b82006-09-27 01:50:18 -0700167
168 return i;
169
170finish_or_fault:
171 return i ? : -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172}
Nick Pigginb291f002008-10-18 20:26:44 -0700173
Nick Pigginb291f002008-10-18 20:26:44 -0700174/*
175 * get a list of pages in an address range belonging to the specified process
176 * and indicate the VMA that covers each page
177 * - this is potentially dodgy as we may end incrementing the page count of a
178 * slab page or a secondary page from a compound page
179 * - don't permit access to VMAs that don't support it, such as I/O mappings
180 */
181int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
Peter Zijlstra9d737772009-06-25 11:58:55 +0200182 unsigned long start, int nr_pages, int write, int force,
Nick Pigginb291f002008-10-18 20:26:44 -0700183 struct page **pages, struct vm_area_struct **vmas)
184{
185 int flags = 0;
186
187 if (write)
Hugh Dickins58fa8792009-09-21 17:03:31 -0700188 flags |= FOLL_WRITE;
Nick Pigginb291f002008-10-18 20:26:44 -0700189 if (force)
Hugh Dickins58fa8792009-09-21 17:03:31 -0700190 flags |= FOLL_FORCE;
Nick Pigginb291f002008-10-18 20:26:44 -0700191
Peter Zijlstra9d737772009-06-25 11:58:55 +0200192 return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas);
Nick Pigginb291f002008-10-18 20:26:44 -0700193}
Greg Ungerer66aa2b42005-09-12 11:18:10 +1000194EXPORT_SYMBOL(get_user_pages);
195
Paul Mundtdfc2f912009-06-26 04:31:57 +0900196/**
197 * follow_pfn - look up PFN at a user virtual address
198 * @vma: memory mapping
199 * @address: user virtual address
200 * @pfn: location to store found PFN
201 *
202 * Only IO mappings and raw PFN mappings are allowed.
203 *
204 * Returns zero and the pfn at @pfn on success, -ve otherwise.
205 */
206int follow_pfn(struct vm_area_struct *vma, unsigned long address,
207 unsigned long *pfn)
208{
209 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
210 return -EINVAL;
211
212 *pfn = address >> PAGE_SHIFT;
213 return 0;
214}
215EXPORT_SYMBOL(follow_pfn);
216
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217DEFINE_RWLOCK(vmlist_lock);
218struct vm_struct *vmlist;
219
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800220void vfree(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221{
222 kfree(addr);
223}
Paul Mundtb5073172007-07-21 04:37:25 -0700224EXPORT_SYMBOL(vfree);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225
Al Virodd0fc662005-10-07 07:46:04 +0100226void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227{
228 /*
Robert P. J. Day85186092007-10-19 23:11:38 +0200229 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
230 * returns only a logical address.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 */
Nick Piggin84097512006-03-22 00:08:34 -0800232 return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233}
Paul Mundtb5073172007-07-21 04:37:25 -0700234EXPORT_SYMBOL(__vmalloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235
Paul Mundtf905bc42008-02-04 22:29:59 -0800236void *vmalloc_user(unsigned long size)
237{
238 void *ret;
239
240 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
241 PAGE_KERNEL);
242 if (ret) {
243 struct vm_area_struct *vma;
244
245 down_write(&current->mm->mmap_sem);
246 vma = find_vma(current->mm, (unsigned long)ret);
247 if (vma)
248 vma->vm_flags |= VM_USERMAP;
249 up_write(&current->mm->mmap_sem);
250 }
251
252 return ret;
253}
254EXPORT_SYMBOL(vmalloc_user);
255
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800256struct page *vmalloc_to_page(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257{
258 return virt_to_page(addr);
259}
Paul Mundtb5073172007-07-21 04:37:25 -0700260EXPORT_SYMBOL(vmalloc_to_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800262unsigned long vmalloc_to_pfn(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263{
264 return page_to_pfn(virt_to_page(addr));
265}
Paul Mundtb5073172007-07-21 04:37:25 -0700266EXPORT_SYMBOL(vmalloc_to_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267
268long vread(char *buf, char *addr, unsigned long count)
269{
270 memcpy(buf, addr, count);
271 return count;
272}
273
274long vwrite(char *buf, char *addr, unsigned long count)
275{
276 /* Don't allow overflow */
277 if ((unsigned long) addr + count < count)
278 count = -(unsigned long) addr;
279
280 memcpy(addr, buf, count);
281 return(count);
282}
283
284/*
285 * vmalloc - allocate virtually continguos memory
286 *
287 * @size: allocation size
288 *
289 * Allocate enough pages to cover @size from the page level
290 * allocator and map them into continguos kernel virtual space.
291 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +0200292 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293 * use __vmalloc() instead.
294 */
295void *vmalloc(unsigned long size)
296{
297 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
298}
Andrew Mortonf6138882006-02-28 16:59:18 -0800299EXPORT_SYMBOL(vmalloc);
300
301void *vmalloc_node(unsigned long size, int node)
302{
303 return vmalloc(size);
304}
305EXPORT_SYMBOL(vmalloc_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306
Paul Mundt1af446e2008-08-04 16:01:47 +0900307#ifndef PAGE_KERNEL_EXEC
308# define PAGE_KERNEL_EXEC PAGE_KERNEL
309#endif
310
311/**
312 * vmalloc_exec - allocate virtually contiguous, executable memory
313 * @size: allocation size
314 *
315 * Kernel-internal function to allocate enough pages to cover @size
316 * the page level allocator and map them into contiguous and
317 * executable kernel virtual space.
318 *
319 * For tight control over page level allocator and protection flags
320 * use __vmalloc() instead.
321 */
322
323void *vmalloc_exec(unsigned long size)
324{
325 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
326}
327
Paul Mundtb5073172007-07-21 04:37:25 -0700328/**
329 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 * @size: allocation size
331 *
332 * Allocate enough 32bit PA addressable pages to cover @size from the
333 * page level allocator and map them into continguos kernel virtual space.
334 */
335void *vmalloc_32(unsigned long size)
336{
337 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
338}
Paul Mundtb5073172007-07-21 04:37:25 -0700339EXPORT_SYMBOL(vmalloc_32);
340
341/**
342 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
343 * @size: allocation size
344 *
345 * The resulting memory area is 32bit addressable and zeroed so it can be
346 * mapped to userspace without leaking data.
Paul Mundtf905bc42008-02-04 22:29:59 -0800347 *
348 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
349 * remap_vmalloc_range() are permissible.
Paul Mundtb5073172007-07-21 04:37:25 -0700350 */
351void *vmalloc_32_user(unsigned long size)
352{
Paul Mundtf905bc42008-02-04 22:29:59 -0800353 /*
354 * We'll have to sort out the ZONE_DMA bits for 64-bit,
355 * but for now this can simply use vmalloc_user() directly.
356 */
357 return vmalloc_user(size);
Paul Mundtb5073172007-07-21 04:37:25 -0700358}
359EXPORT_SYMBOL(vmalloc_32_user);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360
361void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
362{
363 BUG();
364 return NULL;
365}
Paul Mundtb5073172007-07-21 04:37:25 -0700366EXPORT_SYMBOL(vmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800368void vunmap(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369{
370 BUG();
371}
Paul Mundtb5073172007-07-21 04:37:25 -0700372EXPORT_SYMBOL(vunmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373
Paul Mundteb6434d2009-01-21 17:45:47 +0900374void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
375{
376 BUG();
377 return NULL;
378}
379EXPORT_SYMBOL(vm_map_ram);
380
381void vm_unmap_ram(const void *mem, unsigned int count)
382{
383 BUG();
384}
385EXPORT_SYMBOL(vm_unmap_ram);
386
387void vm_unmap_aliases(void)
388{
389}
390EXPORT_SYMBOL_GPL(vm_unmap_aliases);
391
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392/*
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -0700393 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
394 * have one.
395 */
396void __attribute__((weak)) vmalloc_sync_all(void)
397{
398}
399
Paul Mundtb5073172007-07-21 04:37:25 -0700400int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
401 struct page *page)
402{
403 return -EINVAL;
404}
405EXPORT_SYMBOL(vm_insert_page);
406
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -0700407/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 * sys_brk() for the most part doesn't need the global kernel
409 * lock, except when an application is doing something nasty
410 * like trying to un-brk an area that has already been mapped
411 * to a regular file. in this case, the unmapping will need
412 * to invoke file system routines that need the global lock.
413 */
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100414SYSCALL_DEFINE1(brk, unsigned long, brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415{
416 struct mm_struct *mm = current->mm;
417
418 if (brk < mm->start_brk || brk > mm->context.end_brk)
419 return mm->brk;
420
421 if (mm->brk == brk)
422 return mm->brk;
423
424 /*
425 * Always allow shrinking brk
426 */
427 if (brk <= mm->brk) {
428 mm->brk = brk;
429 return brk;
430 }
431
432 /*
433 * Ok, looks good - let it rip.
434 */
Mike Frysingercfe79c02010-01-06 17:23:23 +0000435 flush_icache_range(mm->brk, brk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 return mm->brk = brk;
437}
438
David Howells8feae132009-01-08 12:04:47 +0000439/*
440 * initialise the VMA and region record slabs
441 */
442void __init mmap_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443{
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -0700444 int ret;
445
446 ret = percpu_counter_init(&vm_committed_as, 0);
447 VM_BUG_ON(ret);
David Howells33e5d7692009-04-02 16:56:32 -0700448 vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC);
David Howells8feae132009-01-08 12:04:47 +0000449}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450
David Howells8feae132009-01-08 12:04:47 +0000451/*
452 * validate the region tree
453 * - the caller must hold the region lock
454 */
455#ifdef CONFIG_DEBUG_NOMMU_REGIONS
456static noinline void validate_nommu_regions(void)
457{
458 struct vm_region *region, *last;
459 struct rb_node *p, *lastp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460
David Howells8feae132009-01-08 12:04:47 +0000461 lastp = rb_first(&nommu_region_tree);
462 if (!lastp)
463 return;
464
465 last = rb_entry(lastp, struct vm_region, vm_rb);
David Howells33e5d7692009-04-02 16:56:32 -0700466 BUG_ON(unlikely(last->vm_end <= last->vm_start));
467 BUG_ON(unlikely(last->vm_top < last->vm_end));
David Howells8feae132009-01-08 12:04:47 +0000468
469 while ((p = rb_next(lastp))) {
470 region = rb_entry(p, struct vm_region, vm_rb);
471 last = rb_entry(lastp, struct vm_region, vm_rb);
472
David Howells33e5d7692009-04-02 16:56:32 -0700473 BUG_ON(unlikely(region->vm_end <= region->vm_start));
474 BUG_ON(unlikely(region->vm_top < region->vm_end));
475 BUG_ON(unlikely(region->vm_start < last->vm_top));
David Howells8feae132009-01-08 12:04:47 +0000476
477 lastp = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 }
479}
David Howells8feae132009-01-08 12:04:47 +0000480#else
David Howells33e5d7692009-04-02 16:56:32 -0700481static void validate_nommu_regions(void)
482{
483}
David Howells8feae132009-01-08 12:04:47 +0000484#endif
485
486/*
487 * add a region into the global tree
488 */
489static void add_nommu_region(struct vm_region *region)
490{
491 struct vm_region *pregion;
492 struct rb_node **p, *parent;
493
494 validate_nommu_regions();
495
David Howells8feae132009-01-08 12:04:47 +0000496 parent = NULL;
497 p = &nommu_region_tree.rb_node;
498 while (*p) {
499 parent = *p;
500 pregion = rb_entry(parent, struct vm_region, vm_rb);
501 if (region->vm_start < pregion->vm_start)
502 p = &(*p)->rb_left;
503 else if (region->vm_start > pregion->vm_start)
504 p = &(*p)->rb_right;
505 else if (pregion == region)
506 return;
507 else
508 BUG();
509 }
510
511 rb_link_node(&region->vm_rb, parent, p);
512 rb_insert_color(&region->vm_rb, &nommu_region_tree);
513
514 validate_nommu_regions();
515}
516
517/*
518 * delete a region from the global tree
519 */
520static void delete_nommu_region(struct vm_region *region)
521{
522 BUG_ON(!nommu_region_tree.rb_node);
523
524 validate_nommu_regions();
525 rb_erase(&region->vm_rb, &nommu_region_tree);
526 validate_nommu_regions();
527}
528
529/*
530 * free a contiguous series of pages
531 */
532static void free_page_series(unsigned long from, unsigned long to)
533{
534 for (; from < to; from += PAGE_SIZE) {
535 struct page *page = virt_to_page(from);
536
537 kdebug("- free %lx", from);
David Howells33e5d7692009-04-02 16:56:32 -0700538 atomic_long_dec(&mmap_pages_allocated);
David Howells8feae132009-01-08 12:04:47 +0000539 if (page_count(page) != 1)
David Howells33e5d7692009-04-02 16:56:32 -0700540 kdebug("free page %p: refcount not one: %d",
541 page, page_count(page));
David Howells8feae132009-01-08 12:04:47 +0000542 put_page(page);
543 }
544}
545
546/*
547 * release a reference to a region
David Howells33e5d7692009-04-02 16:56:32 -0700548 * - the caller must hold the region semaphore for writing, which this releases
Paul Mundtdd8632a2009-01-08 12:04:47 +0000549 * - the region may not have been added to the tree yet, in which case vm_top
David Howells8feae132009-01-08 12:04:47 +0000550 * will equal vm_start
551 */
552static void __put_nommu_region(struct vm_region *region)
553 __releases(nommu_region_sem)
554{
David Howells1e2ae592010-01-15 17:01:33 -0800555 kenter("%p{%d}", region, region->vm_usage);
David Howells8feae132009-01-08 12:04:47 +0000556
557 BUG_ON(!nommu_region_tree.rb_node);
558
David Howells1e2ae592010-01-15 17:01:33 -0800559 if (--region->vm_usage == 0) {
Paul Mundtdd8632a2009-01-08 12:04:47 +0000560 if (region->vm_top > region->vm_start)
David Howells8feae132009-01-08 12:04:47 +0000561 delete_nommu_region(region);
562 up_write(&nommu_region_sem);
563
564 if (region->vm_file)
565 fput(region->vm_file);
566
567 /* IO memory and memory shared directly out of the pagecache
568 * from ramfs/tmpfs mustn't be released here */
569 if (region->vm_flags & VM_MAPPED_COPY) {
570 kdebug("free series");
Paul Mundtdd8632a2009-01-08 12:04:47 +0000571 free_page_series(region->vm_start, region->vm_top);
David Howells8feae132009-01-08 12:04:47 +0000572 }
573 kmem_cache_free(vm_region_jar, region);
574 } else {
575 up_write(&nommu_region_sem);
576 }
577}
578
579/*
580 * release a reference to a region
581 */
582static void put_nommu_region(struct vm_region *region)
583{
584 down_write(&nommu_region_sem);
585 __put_nommu_region(region);
586}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587
David Howells30340972006-09-27 01:50:20 -0700588/*
Bernd Schmidteb8cdec2009-09-21 17:03:57 -0700589 * update protection on a vma
590 */
591static void protect_vma(struct vm_area_struct *vma, unsigned long flags)
592{
593#ifdef CONFIG_MPU
594 struct mm_struct *mm = vma->vm_mm;
595 long start = vma->vm_start & PAGE_MASK;
596 while (start < vma->vm_end) {
597 protect_page(mm, start, flags);
598 start += PAGE_SIZE;
599 }
600 update_protections(mm);
601#endif
602}
603
604/*
David Howells30340972006-09-27 01:50:20 -0700605 * add a VMA into a process's mm_struct in the appropriate place in the list
David Howells8feae132009-01-08 12:04:47 +0000606 * and tree and add to the address space's page tree also if not an anonymous
607 * page
David Howells30340972006-09-27 01:50:20 -0700608 * - should be called with mm->mmap_sem held writelocked
609 */
David Howells8feae132009-01-08 12:04:47 +0000610static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
David Howells30340972006-09-27 01:50:20 -0700611{
David Howells8feae132009-01-08 12:04:47 +0000612 struct vm_area_struct *pvma, **pp;
613 struct address_space *mapping;
614 struct rb_node **p, *parent;
David Howells30340972006-09-27 01:50:20 -0700615
David Howells8feae132009-01-08 12:04:47 +0000616 kenter(",%p", vma);
617
618 BUG_ON(!vma->vm_region);
619
620 mm->map_count++;
621 vma->vm_mm = mm;
622
Bernd Schmidteb8cdec2009-09-21 17:03:57 -0700623 protect_vma(vma, vma->vm_flags);
624
David Howells8feae132009-01-08 12:04:47 +0000625 /* add the VMA to the mapping */
626 if (vma->vm_file) {
627 mapping = vma->vm_file->f_mapping;
628
629 flush_dcache_mmap_lock(mapping);
630 vma_prio_tree_insert(vma, &mapping->i_mmap);
631 flush_dcache_mmap_unlock(mapping);
632 }
633
634 /* add the VMA to the tree */
635 parent = NULL;
636 p = &mm->mm_rb.rb_node;
637 while (*p) {
638 parent = *p;
639 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
640
641 /* sort by: start addr, end addr, VMA struct addr in that order
642 * (the latter is necessary as we may get identical VMAs) */
643 if (vma->vm_start < pvma->vm_start)
644 p = &(*p)->rb_left;
645 else if (vma->vm_start > pvma->vm_start)
646 p = &(*p)->rb_right;
647 else if (vma->vm_end < pvma->vm_end)
648 p = &(*p)->rb_left;
649 else if (vma->vm_end > pvma->vm_end)
650 p = &(*p)->rb_right;
651 else if (vma < pvma)
652 p = &(*p)->rb_left;
653 else if (vma > pvma)
654 p = &(*p)->rb_right;
655 else
656 BUG();
657 }
658
659 rb_link_node(&vma->vm_rb, parent, p);
660 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
661
662 /* add VMA to the VMA list also */
663 for (pp = &mm->mmap; (pvma = *pp); pp = &(*pp)->vm_next) {
664 if (pvma->vm_start > vma->vm_start)
David Howells30340972006-09-27 01:50:20 -0700665 break;
David Howells8feae132009-01-08 12:04:47 +0000666 if (pvma->vm_start < vma->vm_start)
667 continue;
668 if (pvma->vm_end < vma->vm_end)
669 break;
670 }
David Howells30340972006-09-27 01:50:20 -0700671
David Howells8feae132009-01-08 12:04:47 +0000672 vma->vm_next = *pp;
673 *pp = vma;
674}
675
676/*
677 * delete a VMA from its owning mm_struct and address space
678 */
679static void delete_vma_from_mm(struct vm_area_struct *vma)
680{
681 struct vm_area_struct **pp;
682 struct address_space *mapping;
683 struct mm_struct *mm = vma->vm_mm;
684
685 kenter("%p", vma);
686
Bernd Schmidteb8cdec2009-09-21 17:03:57 -0700687 protect_vma(vma, 0);
688
David Howells8feae132009-01-08 12:04:47 +0000689 mm->map_count--;
690 if (mm->mmap_cache == vma)
691 mm->mmap_cache = NULL;
692
693 /* remove the VMA from the mapping */
694 if (vma->vm_file) {
695 mapping = vma->vm_file->f_mapping;
696
697 flush_dcache_mmap_lock(mapping);
698 vma_prio_tree_remove(vma, &mapping->i_mmap);
699 flush_dcache_mmap_unlock(mapping);
700 }
701
702 /* remove from the MM's tree and list */
703 rb_erase(&vma->vm_rb, &mm->mm_rb);
704 for (pp = &mm->mmap; *pp; pp = &(*pp)->vm_next) {
705 if (*pp == vma) {
706 *pp = vma->vm_next;
707 break;
708 }
709 }
710
711 vma->vm_mm = NULL;
712}
713
714/*
715 * destroy a VMA record
716 */
717static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
718{
719 kenter("%p", vma);
720 if (vma->vm_ops && vma->vm_ops->close)
721 vma->vm_ops->close(vma);
722 if (vma->vm_file) {
723 fput(vma->vm_file);
724 if (vma->vm_flags & VM_EXECUTABLE)
725 removed_exe_file_vma(mm);
726 }
727 put_nommu_region(vma->vm_region);
728 kmem_cache_free(vm_area_cachep, vma);
David Howells30340972006-09-27 01:50:20 -0700729}
730
731/*
732 * look up the first VMA in which addr resides, NULL if none
733 * - should be called with mm->mmap_sem at least held readlocked
734 */
735struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
736{
David Howells8feae132009-01-08 12:04:47 +0000737 struct vm_area_struct *vma;
738 struct rb_node *n = mm->mm_rb.rb_node;
David Howells30340972006-09-27 01:50:20 -0700739
David Howells8feae132009-01-08 12:04:47 +0000740 /* check the cache first */
741 vma = mm->mmap_cache;
742 if (vma && vma->vm_start <= addr && vma->vm_end > addr)
743 return vma;
744
745 /* trawl the tree (there may be multiple mappings in which addr
746 * resides) */
747 for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) {
748 vma = rb_entry(n, struct vm_area_struct, vm_rb);
749 if (vma->vm_start > addr)
750 return NULL;
751 if (vma->vm_end > addr) {
752 mm->mmap_cache = vma;
753 return vma;
754 }
David Howells30340972006-09-27 01:50:20 -0700755 }
756
David Howells30340972006-09-27 01:50:20 -0700757 return NULL;
758}
759EXPORT_SYMBOL(find_vma);
760
761/*
David Howells930e6522006-09-27 01:50:22 -0700762 * find a VMA
763 * - we don't extend stack VMAs under NOMMU conditions
764 */
765struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
766{
Steven J. Magnanic08c6e12010-03-05 13:42:24 -0800767 return find_vma(mm, addr & PAGE_MASK);
David Howells930e6522006-09-27 01:50:22 -0700768}
769
David Howells8feae132009-01-08 12:04:47 +0000770/*
771 * expand a stack to a given address
772 * - not supported under NOMMU conditions
773 */
Greg Ungerer57c8f632007-07-15 23:38:28 -0700774int expand_stack(struct vm_area_struct *vma, unsigned long address)
775{
776 return -ENOMEM;
777}
778
David Howells930e6522006-09-27 01:50:22 -0700779/*
David Howells6fa5f802006-09-27 01:50:21 -0700780 * look up the first VMA exactly that exactly matches addr
781 * - should be called with mm->mmap_sem at least held readlocked
782 */
David Howells8feae132009-01-08 12:04:47 +0000783static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
784 unsigned long addr,
785 unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786{
787 struct vm_area_struct *vma;
David Howells8feae132009-01-08 12:04:47 +0000788 struct rb_node *n = mm->mm_rb.rb_node;
789 unsigned long end = addr + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790
David Howells8feae132009-01-08 12:04:47 +0000791 /* check the cache first */
792 vma = mm->mmap_cache;
793 if (vma && vma->vm_start == addr && vma->vm_end == end)
794 return vma;
795
796 /* trawl the tree (there may be multiple mappings in which addr
797 * resides) */
798 for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 vma = rb_entry(n, struct vm_area_struct, vm_rb);
David Howells8feae132009-01-08 12:04:47 +0000800 if (vma->vm_start < addr)
801 continue;
802 if (vma->vm_start > addr)
803 return NULL;
804 if (vma->vm_end == end) {
805 mm->mmap_cache = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 return vma;
David Howells8feae132009-01-08 12:04:47 +0000807 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 }
809
810 return NULL;
811}
812
David Howells30340972006-09-27 01:50:20 -0700813/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 * determine whether a mapping should be permitted and, if so, what sort of
815 * mapping we're capable of supporting
816 */
817static int validate_mmap_request(struct file *file,
818 unsigned long addr,
819 unsigned long len,
820 unsigned long prot,
821 unsigned long flags,
822 unsigned long pgoff,
823 unsigned long *_capabilities)
824{
David Howells8feae132009-01-08 12:04:47 +0000825 unsigned long capabilities, rlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 unsigned long reqprot = prot;
827 int ret;
828
829 /* do the simple checks first */
David Howells06aab5a2009-09-24 12:33:48 +0100830 if (flags & MAP_FIXED) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 printk(KERN_DEBUG
832 "%d: Can't do fixed-address/overlay mmap of RAM\n",
833 current->pid);
834 return -EINVAL;
835 }
836
837 if ((flags & MAP_TYPE) != MAP_PRIVATE &&
838 (flags & MAP_TYPE) != MAP_SHARED)
839 return -EINVAL;
840
Mike Frysingerf81cff02006-12-06 12:02:59 +1000841 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 return -EINVAL;
843
Mike Frysingerf81cff02006-12-06 12:02:59 +1000844 /* Careful about overflows.. */
David Howells8feae132009-01-08 12:04:47 +0000845 rlen = PAGE_ALIGN(len);
846 if (!rlen || rlen > TASK_SIZE)
Mike Frysingerf81cff02006-12-06 12:02:59 +1000847 return -ENOMEM;
848
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 /* offset overflow? */
David Howells8feae132009-01-08 12:04:47 +0000850 if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
Mike Frysingerf81cff02006-12-06 12:02:59 +1000851 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852
853 if (file) {
854 /* validate file mapping requests */
855 struct address_space *mapping;
856
857 /* files must support mmap */
858 if (!file->f_op || !file->f_op->mmap)
859 return -ENODEV;
860
861 /* work out if what we've got could possibly be shared
862 * - we support chardevs that provide their own "memory"
863 * - we support files/blockdevs that are memory backed
864 */
865 mapping = file->f_mapping;
866 if (!mapping)
Josef Sipeke9536ae2006-12-08 02:37:21 -0800867 mapping = file->f_path.dentry->d_inode->i_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868
869 capabilities = 0;
870 if (mapping && mapping->backing_dev_info)
871 capabilities = mapping->backing_dev_info->capabilities;
872
873 if (!capabilities) {
874 /* no explicit capabilities set, so assume some
875 * defaults */
Josef Sipeke9536ae2006-12-08 02:37:21 -0800876 switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 case S_IFREG:
878 case S_IFBLK:
879 capabilities = BDI_CAP_MAP_COPY;
880 break;
881
882 case S_IFCHR:
883 capabilities =
884 BDI_CAP_MAP_DIRECT |
885 BDI_CAP_READ_MAP |
886 BDI_CAP_WRITE_MAP;
887 break;
888
889 default:
890 return -EINVAL;
891 }
892 }
893
894 /* eliminate any capabilities that we can't support on this
895 * device */
896 if (!file->f_op->get_unmapped_area)
897 capabilities &= ~BDI_CAP_MAP_DIRECT;
898 if (!file->f_op->read)
899 capabilities &= ~BDI_CAP_MAP_COPY;
900
Graff Yang28d7a6a2009-08-18 14:11:17 -0700901 /* The file shall have been opened with read permission. */
902 if (!(file->f_mode & FMODE_READ))
903 return -EACCES;
904
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 if (flags & MAP_SHARED) {
906 /* do checks for writing, appending and locking */
907 if ((prot & PROT_WRITE) &&
908 !(file->f_mode & FMODE_WRITE))
909 return -EACCES;
910
Josef Sipeke9536ae2006-12-08 02:37:21 -0800911 if (IS_APPEND(file->f_path.dentry->d_inode) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 (file->f_mode & FMODE_WRITE))
913 return -EACCES;
914
Josef Sipeke9536ae2006-12-08 02:37:21 -0800915 if (locks_verify_locked(file->f_path.dentry->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 return -EAGAIN;
917
918 if (!(capabilities & BDI_CAP_MAP_DIRECT))
919 return -ENODEV;
920
921 if (((prot & PROT_READ) && !(capabilities & BDI_CAP_READ_MAP)) ||
922 ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
923 ((prot & PROT_EXEC) && !(capabilities & BDI_CAP_EXEC_MAP))
924 ) {
925 printk("MAP_SHARED not completely supported on !MMU\n");
926 return -EINVAL;
927 }
928
929 /* we mustn't privatise shared mappings */
930 capabilities &= ~BDI_CAP_MAP_COPY;
931 }
932 else {
933 /* we're going to read the file into private memory we
934 * allocate */
935 if (!(capabilities & BDI_CAP_MAP_COPY))
936 return -ENODEV;
937
938 /* we don't permit a private writable mapping to be
939 * shared with the backing device */
940 if (prot & PROT_WRITE)
941 capabilities &= ~BDI_CAP_MAP_DIRECT;
942 }
943
944 /* handle executable mappings and implied executable
945 * mappings */
Josef Sipeke9536ae2006-12-08 02:37:21 -0800946 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 if (prot & PROT_EXEC)
948 return -EPERM;
949 }
950 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
951 /* handle implication of PROT_EXEC by PROT_READ */
952 if (current->personality & READ_IMPLIES_EXEC) {
953 if (capabilities & BDI_CAP_EXEC_MAP)
954 prot |= PROT_EXEC;
955 }
956 }
957 else if ((prot & PROT_READ) &&
958 (prot & PROT_EXEC) &&
959 !(capabilities & BDI_CAP_EXEC_MAP)
960 ) {
961 /* backing file is not executable, try to copy */
962 capabilities &= ~BDI_CAP_MAP_DIRECT;
963 }
964 }
965 else {
966 /* anonymous mappings are always memory backed and can be
967 * privately mapped
968 */
969 capabilities = BDI_CAP_MAP_COPY;
970
971 /* handle PROT_EXEC implication by PROT_READ */
972 if ((prot & PROT_READ) &&
973 (current->personality & READ_IMPLIES_EXEC))
974 prot |= PROT_EXEC;
975 }
976
977 /* allow the security API to have its say */
Eric Parised032182007-06-28 15:55:21 -0400978 ret = security_file_mmap(file, reqprot, prot, flags, addr, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 if (ret < 0)
980 return ret;
981
982 /* looks okay */
983 *_capabilities = capabilities;
984 return 0;
985}
986
987/*
988 * we've determined that we can make the mapping, now translate what we
989 * now know into VMA flags
990 */
991static unsigned long determine_vm_flags(struct file *file,
992 unsigned long prot,
993 unsigned long flags,
994 unsigned long capabilities)
995{
996 unsigned long vm_flags;
997
998 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
999 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1000 /* vm_flags |= mm->def_flags; */
1001
1002 if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
1003 /* attempt to share read-only copies of mapped file chunks */
1004 if (file && !(prot & PROT_WRITE))
1005 vm_flags |= VM_MAYSHARE;
1006 }
1007 else {
1008 /* overlay a shareable mapping on the backing device or inode
1009 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1010 * romfs/cramfs */
1011 if (flags & MAP_SHARED)
1012 vm_flags |= VM_MAYSHARE | VM_SHARED;
1013 else if ((((vm_flags & capabilities) ^ vm_flags) & BDI_CAP_VMFLAGS) == 0)
1014 vm_flags |= VM_MAYSHARE;
1015 }
1016
1017 /* refuse to let anyone share private mappings with this process if
1018 * it's being traced - otherwise breakpoints set in it may interfere
1019 * with another untraced process
1020 */
Roland McGrathfa8e26c2008-07-25 19:45:50 -07001021 if ((flags & MAP_PRIVATE) && tracehook_expect_breakpoints(current))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 vm_flags &= ~VM_MAYSHARE;
1023
1024 return vm_flags;
1025}
1026
1027/*
David Howells8feae132009-01-08 12:04:47 +00001028 * set up a shared mapping on a file (the driver or filesystem provides and
1029 * pins the storage)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 */
David Howells8feae132009-01-08 12:04:47 +00001031static int do_mmap_shared_file(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032{
1033 int ret;
1034
1035 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001036 if (ret == 0) {
1037 vma->vm_region->vm_top = vma->vm_region->vm_end;
David Howells645d83c2009-09-24 15:13:10 +01001038 return 0;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001039 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 if (ret != -ENOSYS)
1041 return ret;
1042
1043 /* getting an ENOSYS error indicates that direct mmap isn't
1044 * possible (as opposed to tried but failed) so we'll fall
1045 * through to making a private copy of the data and mapping
1046 * that if we can */
1047 return -ENODEV;
1048}
1049
1050/*
1051 * set up a private mapping or an anonymous shared mapping
1052 */
David Howells8feae132009-01-08 12:04:47 +00001053static int do_mmap_private(struct vm_area_struct *vma,
1054 struct vm_region *region,
David Howells645d83c2009-09-24 15:13:10 +01001055 unsigned long len,
1056 unsigned long capabilities)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057{
David Howells8feae132009-01-08 12:04:47 +00001058 struct page *pages;
1059 unsigned long total, point, n, rlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 void *base;
David Howells8feae132009-01-08 12:04:47 +00001061 int ret, order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062
1063 /* invoke the file's mapping function so that it can keep track of
1064 * shared mappings on devices or memory
1065 * - VM_MAYSHARE will be set if it may attempt to share
1066 */
David Howells645d83c2009-09-24 15:13:10 +01001067 if (capabilities & BDI_CAP_MAP_DIRECT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001069 if (ret == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 /* shouldn't return success if we're not sharing */
Paul Mundtdd8632a2009-01-08 12:04:47 +00001071 BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1072 vma->vm_region->vm_top = vma->vm_region->vm_end;
David Howells645d83c2009-09-24 15:13:10 +01001073 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 }
Paul Mundtdd8632a2009-01-08 12:04:47 +00001075 if (ret != -ENOSYS)
1076 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077
1078 /* getting an ENOSYS error indicates that direct mmap isn't
1079 * possible (as opposed to tried but failed) so we'll try to
1080 * make a private copy of the data and map that instead */
1081 }
1082
David Howells8feae132009-01-08 12:04:47 +00001083 rlen = PAGE_ALIGN(len);
1084
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 /* allocate some memory to hold the mapping
1086 * - note that this may not return a page-aligned address if the object
1087 * we're allocating is smaller than a page
1088 */
David Howells8feae132009-01-08 12:04:47 +00001089 order = get_order(rlen);
1090 kdebug("alloc order %d for %lx", order, len);
1091
1092 pages = alloc_pages(GFP_KERNEL, order);
1093 if (!pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 goto enomem;
1095
David Howells8feae132009-01-08 12:04:47 +00001096 total = 1 << order;
David Howells33e5d7692009-04-02 16:56:32 -07001097 atomic_long_add(total, &mmap_pages_allocated);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098
David Howells8feae132009-01-08 12:04:47 +00001099 point = rlen >> PAGE_SHIFT;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001100
1101 /* we allocated a power-of-2 sized page set, so we may want to trim off
1102 * the excess */
1103 if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) {
1104 while (total > point) {
1105 order = ilog2(total - point);
1106 n = 1 << order;
1107 kdebug("shave %lu/%lu @%lu", n, total - point, total);
David Howells33e5d7692009-04-02 16:56:32 -07001108 atomic_long_sub(n, &mmap_pages_allocated);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001109 total -= n;
1110 set_page_refcounted(pages + total);
1111 __free_pages(pages + total, order);
1112 }
David Howells8feae132009-01-08 12:04:47 +00001113 }
1114
David Howells8feae132009-01-08 12:04:47 +00001115 for (point = 1; point < total; point++)
1116 set_page_refcounted(&pages[point]);
1117
1118 base = page_address(pages);
1119 region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1120 region->vm_start = (unsigned long) base;
1121 region->vm_end = region->vm_start + rlen;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001122 region->vm_top = region->vm_start + (total << PAGE_SHIFT);
David Howells8feae132009-01-08 12:04:47 +00001123
1124 vma->vm_start = region->vm_start;
1125 vma->vm_end = region->vm_start + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126
1127 if (vma->vm_file) {
1128 /* read the contents of a file into the copy */
1129 mm_segment_t old_fs;
1130 loff_t fpos;
1131
1132 fpos = vma->vm_pgoff;
1133 fpos <<= PAGE_SHIFT;
1134
1135 old_fs = get_fs();
1136 set_fs(KERNEL_DS);
David Howells8feae132009-01-08 12:04:47 +00001137 ret = vma->vm_file->f_op->read(vma->vm_file, base, rlen, &fpos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138 set_fs(old_fs);
1139
1140 if (ret < 0)
1141 goto error_free;
1142
1143 /* clear the last little bit */
David Howells8feae132009-01-08 12:04:47 +00001144 if (ret < rlen)
1145 memset(base + ret, 0, rlen - ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 }
1148
1149 return 0;
1150
1151error_free:
David Howells8feae132009-01-08 12:04:47 +00001152 free_page_series(region->vm_start, region->vm_end);
1153 region->vm_start = vma->vm_start = 0;
1154 region->vm_end = vma->vm_end = 0;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001155 region->vm_top = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 return ret;
1157
1158enomem:
Greg Ungerer05ae6fa2009-01-13 17:30:22 +10001159 printk("Allocation of length %lu from process %d (%s) failed\n",
1160 len, current->pid, current->comm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 show_free_areas();
1162 return -ENOMEM;
1163}
1164
1165/*
1166 * handle mapping creation for uClinux
1167 */
1168unsigned long do_mmap_pgoff(struct file *file,
1169 unsigned long addr,
1170 unsigned long len,
1171 unsigned long prot,
1172 unsigned long flags,
1173 unsigned long pgoff)
1174{
David Howells8feae132009-01-08 12:04:47 +00001175 struct vm_area_struct *vma;
1176 struct vm_region *region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 struct rb_node *rb;
David Howells8feae132009-01-08 12:04:47 +00001178 unsigned long capabilities, vm_flags, result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 int ret;
1180
David Howells8feae132009-01-08 12:04:47 +00001181 kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff);
1182
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183 /* decide whether we should attempt the mapping, and if so what sort of
1184 * mapping */
1185 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1186 &capabilities);
David Howells8feae132009-01-08 12:04:47 +00001187 if (ret < 0) {
1188 kleave(" = %d [val]", ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189 return ret;
David Howells8feae132009-01-08 12:04:47 +00001190 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191
David Howells06aab5a2009-09-24 12:33:48 +01001192 /* we ignore the address hint */
1193 addr = 0;
1194
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 /* we've determined that we can make the mapping, now translate what we
1196 * now know into VMA flags */
1197 vm_flags = determine_vm_flags(file, prot, flags, capabilities);
1198
David Howells8feae132009-01-08 12:04:47 +00001199 /* we're going to need to record the mapping */
1200 region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1201 if (!region)
1202 goto error_getting_region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203
David Howells8feae132009-01-08 12:04:47 +00001204 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1205 if (!vma)
1206 goto error_getting_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207
David Howells1e2ae592010-01-15 17:01:33 -08001208 region->vm_usage = 1;
David Howells8feae132009-01-08 12:04:47 +00001209 region->vm_flags = vm_flags;
1210 region->vm_pgoff = pgoff;
1211
Rik van Riel5beb4932010-03-05 13:42:07 -08001212 INIT_LIST_HEAD(&vma->anon_vma_chain);
David Howells8feae132009-01-08 12:04:47 +00001213 vma->vm_flags = vm_flags;
1214 vma->vm_pgoff = pgoff;
1215
1216 if (file) {
1217 region->vm_file = file;
1218 get_file(file);
1219 vma->vm_file = file;
1220 get_file(file);
1221 if (vm_flags & VM_EXECUTABLE) {
1222 added_exe_file_vma(current->mm);
1223 vma->vm_mm = current->mm;
1224 }
1225 }
1226
1227 down_write(&nommu_region_sem);
1228
1229 /* if we want to share, we need to check for regions created by other
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 * mmap() calls that overlap with our proposed mapping
David Howells8feae132009-01-08 12:04:47 +00001231 * - we can only share with a superset match on most regular files
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 * - shared mappings on character devices and memory backed files are
1233 * permitted to overlap inexactly as far as we are concerned for in
1234 * these cases, sharing is handled in the driver or filesystem rather
1235 * than here
1236 */
1237 if (vm_flags & VM_MAYSHARE) {
David Howells8feae132009-01-08 12:04:47 +00001238 struct vm_region *pregion;
1239 unsigned long pglen, rpglen, pgend, rpgend, start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240
David Howells8feae132009-01-08 12:04:47 +00001241 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1242 pgend = pgoff + pglen;
David Howells165b2392007-03-22 00:11:24 -08001243
David Howells8feae132009-01-08 12:04:47 +00001244 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1245 pregion = rb_entry(rb, struct vm_region, vm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246
David Howells8feae132009-01-08 12:04:47 +00001247 if (!(pregion->vm_flags & VM_MAYSHARE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248 continue;
1249
1250 /* search for overlapping mappings on the same file */
David Howells8feae132009-01-08 12:04:47 +00001251 if (pregion->vm_file->f_path.dentry->d_inode !=
1252 file->f_path.dentry->d_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 continue;
1254
David Howells8feae132009-01-08 12:04:47 +00001255 if (pregion->vm_pgoff >= pgend)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 continue;
1257
David Howells8feae132009-01-08 12:04:47 +00001258 rpglen = pregion->vm_end - pregion->vm_start;
1259 rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1260 rpgend = pregion->vm_pgoff + rpglen;
1261 if (pgoff >= rpgend)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 continue;
1263
David Howells8feae132009-01-08 12:04:47 +00001264 /* handle inexactly overlapping matches between
1265 * mappings */
1266 if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1267 !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1268 /* new mapping is not a subset of the region */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 if (!(capabilities & BDI_CAP_MAP_DIRECT))
1270 goto sharing_violation;
1271 continue;
1272 }
1273
David Howells8feae132009-01-08 12:04:47 +00001274 /* we've found a region we can share */
David Howells1e2ae592010-01-15 17:01:33 -08001275 pregion->vm_usage++;
David Howells8feae132009-01-08 12:04:47 +00001276 vma->vm_region = pregion;
1277 start = pregion->vm_start;
1278 start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1279 vma->vm_start = start;
1280 vma->vm_end = start + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281
David Howells8feae132009-01-08 12:04:47 +00001282 if (pregion->vm_flags & VM_MAPPED_COPY) {
1283 kdebug("share copy");
1284 vma->vm_flags |= VM_MAPPED_COPY;
1285 } else {
1286 kdebug("share mmap");
1287 ret = do_mmap_shared_file(vma);
1288 if (ret < 0) {
1289 vma->vm_region = NULL;
1290 vma->vm_start = 0;
1291 vma->vm_end = 0;
David Howells1e2ae592010-01-15 17:01:33 -08001292 pregion->vm_usage--;
David Howells8feae132009-01-08 12:04:47 +00001293 pregion = NULL;
1294 goto error_just_free;
1295 }
1296 }
1297 fput(region->vm_file);
1298 kmem_cache_free(vm_region_jar, region);
1299 region = pregion;
1300 result = start;
1301 goto share;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302 }
1303
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 /* obtain the address at which to make a shared mapping
1305 * - this is the hook for quasi-memory character devices to
1306 * tell us the location of a shared mapping
1307 */
David Howells645d83c2009-09-24 15:13:10 +01001308 if (capabilities & BDI_CAP_MAP_DIRECT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 addr = file->f_op->get_unmapped_area(file, addr, len,
1310 pgoff, flags);
1311 if (IS_ERR((void *) addr)) {
1312 ret = addr;
1313 if (ret != (unsigned long) -ENOSYS)
David Howells8feae132009-01-08 12:04:47 +00001314 goto error_just_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315
1316 /* the driver refused to tell us where to site
1317 * the mapping so we'll have to attempt to copy
1318 * it */
1319 ret = (unsigned long) -ENODEV;
1320 if (!(capabilities & BDI_CAP_MAP_COPY))
David Howells8feae132009-01-08 12:04:47 +00001321 goto error_just_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322
1323 capabilities &= ~BDI_CAP_MAP_DIRECT;
David Howells8feae132009-01-08 12:04:47 +00001324 } else {
1325 vma->vm_start = region->vm_start = addr;
1326 vma->vm_end = region->vm_end = addr + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327 }
1328 }
1329 }
1330
David Howells8feae132009-01-08 12:04:47 +00001331 vma->vm_region = region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332
David Howells645d83c2009-09-24 15:13:10 +01001333 /* set up the mapping
1334 * - the region is filled in if BDI_CAP_MAP_DIRECT is still set
1335 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 if (file && vma->vm_flags & VM_SHARED)
David Howells8feae132009-01-08 12:04:47 +00001337 ret = do_mmap_shared_file(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 else
David Howells645d83c2009-09-24 15:13:10 +01001339 ret = do_mmap_private(vma, region, len, capabilities);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 if (ret < 0)
David Howells645d83c2009-09-24 15:13:10 +01001341 goto error_just_free;
1342 add_nommu_region(region);
David Howells8feae132009-01-08 12:04:47 +00001343
Jie Zhangea637632009-12-14 18:00:02 -08001344 /* clear anonymous mappings that don't ask for uninitialized data */
1345 if (!vma->vm_file && !(flags & MAP_UNINITIALIZED))
1346 memset((void *)region->vm_start, 0,
1347 region->vm_end - region->vm_start);
1348
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 /* okay... we have a mapping; now we have to register it */
David Howells8feae132009-01-08 12:04:47 +00001350 result = vma->vm_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 current->mm->total_vm += len >> PAGE_SHIFT;
1353
David Howells8feae132009-01-08 12:04:47 +00001354share:
1355 add_vma_to_mm(current->mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356
Mike Frysingercfe79c02010-01-06 17:23:23 +00001357 /* we flush the region from the icache only when the first executable
1358 * mapping of it is made */
1359 if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1360 flush_icache_range(region->vm_start, region->vm_end);
1361 region->vm_icache_flushed = true;
1362 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363
Mike Frysingercfe79c02010-01-06 17:23:23 +00001364 up_write(&nommu_region_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365
David Howells8feae132009-01-08 12:04:47 +00001366 kleave(" = %lx", result);
1367 return result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368
David Howells8feae132009-01-08 12:04:47 +00001369error_just_free:
1370 up_write(&nommu_region_sem);
1371error:
David Howells89a86402009-10-30 13:13:26 +00001372 if (region->vm_file)
1373 fput(region->vm_file);
David Howells8feae132009-01-08 12:04:47 +00001374 kmem_cache_free(vm_region_jar, region);
David Howells89a86402009-10-30 13:13:26 +00001375 if (vma->vm_file)
1376 fput(vma->vm_file);
David Howells8feae132009-01-08 12:04:47 +00001377 if (vma->vm_flags & VM_EXECUTABLE)
1378 removed_exe_file_vma(vma->vm_mm);
1379 kmem_cache_free(vm_area_cachep, vma);
1380 kleave(" = %d", ret);
1381 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382
David Howells8feae132009-01-08 12:04:47 +00001383sharing_violation:
1384 up_write(&nommu_region_sem);
1385 printk(KERN_WARNING "Attempt to share mismatched mappings\n");
1386 ret = -EINVAL;
1387 goto error;
1388
1389error_getting_vma:
1390 kmem_cache_free(vm_region_jar, region);
1391 printk(KERN_WARNING "Allocation of vma for %lu byte allocation"
1392 " from process %d failed\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393 len, current->pid);
1394 show_free_areas();
1395 return -ENOMEM;
1396
David Howells8feae132009-01-08 12:04:47 +00001397error_getting_region:
1398 printk(KERN_WARNING "Allocation of vm region for %lu byte allocation"
1399 " from process %d failed\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400 len, current->pid);
1401 show_free_areas();
1402 return -ENOMEM;
1403}
Paul Mundtb5073172007-07-21 04:37:25 -07001404EXPORT_SYMBOL(do_mmap_pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405
Hugh Dickins66f0dc42009-12-30 20:17:34 +00001406SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1407 unsigned long, prot, unsigned long, flags,
1408 unsigned long, fd, unsigned long, pgoff)
1409{
1410 struct file *file = NULL;
1411 unsigned long retval = -EBADF;
1412
1413 if (!(flags & MAP_ANONYMOUS)) {
1414 file = fget(fd);
1415 if (!file)
1416 goto out;
1417 }
1418
1419 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1420
1421 down_write(&current->mm->mmap_sem);
1422 retval = do_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1423 up_write(&current->mm->mmap_sem);
1424
1425 if (file)
1426 fput(file);
1427out:
1428 return retval;
1429}
1430
Christoph Hellwiga4679372010-03-10 15:21:15 -08001431#ifdef __ARCH_WANT_SYS_OLD_MMAP
1432struct mmap_arg_struct {
1433 unsigned long addr;
1434 unsigned long len;
1435 unsigned long prot;
1436 unsigned long flags;
1437 unsigned long fd;
1438 unsigned long offset;
1439};
1440
1441SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1442{
1443 struct mmap_arg_struct a;
1444
1445 if (copy_from_user(&a, arg, sizeof(a)))
1446 return -EFAULT;
1447 if (a.offset & ~PAGE_MASK)
1448 return -EINVAL;
1449
1450 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1451 a.offset >> PAGE_SHIFT);
1452}
1453#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1454
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455/*
David Howells8feae132009-01-08 12:04:47 +00001456 * split a vma into two pieces at address 'addr', a new vma is allocated either
1457 * for the first part or the tail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 */
David Howells8feae132009-01-08 12:04:47 +00001459int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1460 unsigned long addr, int new_below)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461{
David Howells8feae132009-01-08 12:04:47 +00001462 struct vm_area_struct *new;
1463 struct vm_region *region;
1464 unsigned long npages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465
David Howells8feae132009-01-08 12:04:47 +00001466 kenter("");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467
David Howells779c1022010-01-15 17:01:34 -08001468 /* we're only permitted to split anonymous regions (these should have
1469 * only a single usage on the region) */
1470 if (vma->vm_file)
David Howells8feae132009-01-08 12:04:47 +00001471 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472
David Howells8feae132009-01-08 12:04:47 +00001473 if (mm->map_count >= sysctl_max_map_count)
1474 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475
David Howells8feae132009-01-08 12:04:47 +00001476 region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1477 if (!region)
1478 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479
David Howells8feae132009-01-08 12:04:47 +00001480 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1481 if (!new) {
1482 kmem_cache_free(vm_region_jar, region);
1483 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 }
David Howells8feae132009-01-08 12:04:47 +00001485
1486 /* most fields are the same, copy all, and then fixup */
1487 *new = *vma;
1488 *region = *vma->vm_region;
1489 new->vm_region = region;
1490
1491 npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1492
1493 if (new_below) {
Paul Mundtdd8632a2009-01-08 12:04:47 +00001494 region->vm_top = region->vm_end = new->vm_end = addr;
David Howells8feae132009-01-08 12:04:47 +00001495 } else {
1496 region->vm_start = new->vm_start = addr;
1497 region->vm_pgoff = new->vm_pgoff += npages;
1498 }
1499
1500 if (new->vm_ops && new->vm_ops->open)
1501 new->vm_ops->open(new);
1502
1503 delete_vma_from_mm(vma);
1504 down_write(&nommu_region_sem);
1505 delete_nommu_region(vma->vm_region);
1506 if (new_below) {
1507 vma->vm_region->vm_start = vma->vm_start = addr;
1508 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1509 } else {
1510 vma->vm_region->vm_end = vma->vm_end = addr;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001511 vma->vm_region->vm_top = addr;
David Howells8feae132009-01-08 12:04:47 +00001512 }
1513 add_nommu_region(vma->vm_region);
1514 add_nommu_region(new->vm_region);
1515 up_write(&nommu_region_sem);
1516 add_vma_to_mm(mm, vma);
1517 add_vma_to_mm(mm, new);
1518 return 0;
1519}
1520
1521/*
1522 * shrink a VMA by removing the specified chunk from either the beginning or
1523 * the end
1524 */
1525static int shrink_vma(struct mm_struct *mm,
1526 struct vm_area_struct *vma,
1527 unsigned long from, unsigned long to)
1528{
1529 struct vm_region *region;
1530
1531 kenter("");
1532
1533 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1534 * and list */
1535 delete_vma_from_mm(vma);
1536 if (from > vma->vm_start)
1537 vma->vm_end = from;
1538 else
1539 vma->vm_start = to;
1540 add_vma_to_mm(mm, vma);
1541
1542 /* cut the backing region down to size */
1543 region = vma->vm_region;
David Howells1e2ae592010-01-15 17:01:33 -08001544 BUG_ON(region->vm_usage != 1);
David Howells8feae132009-01-08 12:04:47 +00001545
1546 down_write(&nommu_region_sem);
1547 delete_nommu_region(region);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001548 if (from > region->vm_start) {
1549 to = region->vm_top;
1550 region->vm_top = region->vm_end = from;
1551 } else {
David Howells8feae132009-01-08 12:04:47 +00001552 region->vm_start = to;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001553 }
David Howells8feae132009-01-08 12:04:47 +00001554 add_nommu_region(region);
1555 up_write(&nommu_region_sem);
1556
1557 free_page_series(from, to);
1558 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559}
1560
David Howells30340972006-09-27 01:50:20 -07001561/*
1562 * release a mapping
David Howells8feae132009-01-08 12:04:47 +00001563 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1564 * VMA, though it need not cover the whole VMA
David Howells30340972006-09-27 01:50:20 -07001565 */
David Howells8feae132009-01-08 12:04:47 +00001566int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567{
David Howells8feae132009-01-08 12:04:47 +00001568 struct vm_area_struct *vma;
1569 struct rb_node *rb;
1570 unsigned long end = start + len;
1571 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572
David Howells8feae132009-01-08 12:04:47 +00001573 kenter(",%lx,%zx", start, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574
David Howells8feae132009-01-08 12:04:47 +00001575 if (len == 0)
1576 return -EINVAL;
1577
1578 /* find the first potentially overlapping VMA */
1579 vma = find_vma(mm, start);
1580 if (!vma) {
David Howells33e5d7692009-04-02 16:56:32 -07001581 static int limit = 0;
1582 if (limit < 5) {
1583 printk(KERN_WARNING
1584 "munmap of memory not mmapped by process %d"
1585 " (%s): 0x%lx-0x%lx\n",
1586 current->pid, current->comm,
1587 start, start + len - 1);
1588 limit++;
1589 }
David Howells8feae132009-01-08 12:04:47 +00001590 return -EINVAL;
David Howells30340972006-09-27 01:50:20 -07001591 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592
David Howells8feae132009-01-08 12:04:47 +00001593 /* we're allowed to split an anonymous VMA but not a file-backed one */
1594 if (vma->vm_file) {
1595 do {
1596 if (start > vma->vm_start) {
1597 kleave(" = -EINVAL [miss]");
1598 return -EINVAL;
1599 }
1600 if (end == vma->vm_end)
1601 goto erase_whole_vma;
1602 rb = rb_next(&vma->vm_rb);
1603 vma = rb_entry(rb, struct vm_area_struct, vm_rb);
1604 } while (rb);
1605 kleave(" = -EINVAL [split file]");
1606 return -EINVAL;
1607 } else {
1608 /* the chunk must be a subset of the VMA found */
1609 if (start == vma->vm_start && end == vma->vm_end)
1610 goto erase_whole_vma;
1611 if (start < vma->vm_start || end > vma->vm_end) {
1612 kleave(" = -EINVAL [superset]");
1613 return -EINVAL;
1614 }
1615 if (start & ~PAGE_MASK) {
1616 kleave(" = -EINVAL [unaligned start]");
1617 return -EINVAL;
1618 }
1619 if (end != vma->vm_end && end & ~PAGE_MASK) {
1620 kleave(" = -EINVAL [unaligned split]");
1621 return -EINVAL;
1622 }
1623 if (start != vma->vm_start && end != vma->vm_end) {
1624 ret = split_vma(mm, vma, start, 1);
1625 if (ret < 0) {
1626 kleave(" = %d [split]", ret);
1627 return ret;
1628 }
1629 }
1630 return shrink_vma(mm, vma, start, end);
1631 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632
David Howells8feae132009-01-08 12:04:47 +00001633erase_whole_vma:
1634 delete_vma_from_mm(vma);
1635 delete_vma(mm, vma);
1636 kleave(" = 0");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 return 0;
1638}
Paul Mundtb5073172007-07-21 04:37:25 -07001639EXPORT_SYMBOL(do_munmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640
Heiko Carstens6a6160a2009-01-14 14:14:15 +01001641SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
David Howells30340972006-09-27 01:50:20 -07001642{
1643 int ret;
1644 struct mm_struct *mm = current->mm;
1645
1646 down_write(&mm->mmap_sem);
1647 ret = do_munmap(mm, addr, len);
1648 up_write(&mm->mmap_sem);
1649 return ret;
1650}
1651
1652/*
David Howells8feae132009-01-08 12:04:47 +00001653 * release all the mappings made in a process's VM space
David Howells30340972006-09-27 01:50:20 -07001654 */
David Howells8feae132009-01-08 12:04:47 +00001655void exit_mmap(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656{
David Howells8feae132009-01-08 12:04:47 +00001657 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658
David Howells8feae132009-01-08 12:04:47 +00001659 if (!mm)
1660 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661
David Howells8feae132009-01-08 12:04:47 +00001662 kenter("");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663
David Howells8feae132009-01-08 12:04:47 +00001664 mm->total_vm = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665
David Howells8feae132009-01-08 12:04:47 +00001666 while ((vma = mm->mmap)) {
1667 mm->mmap = vma->vm_next;
1668 delete_vma_from_mm(vma);
1669 delete_vma(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 }
David Howells8feae132009-01-08 12:04:47 +00001671
1672 kleave("");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673}
1674
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675unsigned long do_brk(unsigned long addr, unsigned long len)
1676{
1677 return -ENOMEM;
1678}
1679
1680/*
David Howells6fa5f802006-09-27 01:50:21 -07001681 * expand (or shrink) an existing mapping, potentially moving it at the same
1682 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683 *
David Howells6fa5f802006-09-27 01:50:21 -07001684 * under NOMMU conditions, we only permit changing a mapping's size, and only
David Howells8feae132009-01-08 12:04:47 +00001685 * as long as it stays within the region allocated by do_mmap_private() and the
1686 * block is not shareable
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687 *
David Howells6fa5f802006-09-27 01:50:21 -07001688 * MREMAP_FIXED is not supported under NOMMU conditions
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689 */
1690unsigned long do_mremap(unsigned long addr,
1691 unsigned long old_len, unsigned long new_len,
1692 unsigned long flags, unsigned long new_addr)
1693{
David Howells6fa5f802006-09-27 01:50:21 -07001694 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695
1696 /* insanity checks first */
David Howells8feae132009-01-08 12:04:47 +00001697 if (old_len == 0 || new_len == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698 return (unsigned long) -EINVAL;
1699
David Howells8feae132009-01-08 12:04:47 +00001700 if (addr & ~PAGE_MASK)
1701 return -EINVAL;
1702
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703 if (flags & MREMAP_FIXED && new_addr != addr)
1704 return (unsigned long) -EINVAL;
1705
David Howells8feae132009-01-08 12:04:47 +00001706 vma = find_vma_exact(current->mm, addr, old_len);
David Howells6fa5f802006-09-27 01:50:21 -07001707 if (!vma)
1708 return (unsigned long) -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709
David Howells6fa5f802006-09-27 01:50:21 -07001710 if (vma->vm_end != vma->vm_start + old_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711 return (unsigned long) -EFAULT;
1712
David Howells6fa5f802006-09-27 01:50:21 -07001713 if (vma->vm_flags & VM_MAYSHARE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714 return (unsigned long) -EPERM;
1715
David Howells8feae132009-01-08 12:04:47 +00001716 if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717 return (unsigned long) -ENOMEM;
1718
1719 /* all checks complete - do it */
David Howells6fa5f802006-09-27 01:50:21 -07001720 vma->vm_end = vma->vm_start + new_len;
David Howells6fa5f802006-09-27 01:50:21 -07001721 return vma->vm_start;
1722}
Paul Mundtb5073172007-07-21 04:37:25 -07001723EXPORT_SYMBOL(do_mremap);
David Howells6fa5f802006-09-27 01:50:21 -07001724
Heiko Carstens6a6160a2009-01-14 14:14:15 +01001725SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1726 unsigned long, new_len, unsigned long, flags,
1727 unsigned long, new_addr)
David Howells6fa5f802006-09-27 01:50:21 -07001728{
1729 unsigned long ret;
1730
1731 down_write(&current->mm->mmap_sem);
1732 ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1733 up_write(&current->mm->mmap_sem);
1734 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735}
1736
Linus Torvalds6aab3412005-11-28 14:34:23 -08001737struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001738 unsigned int foll_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739{
1740 return NULL;
1741}
1742
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743int remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
1744 unsigned long to, unsigned long size, pgprot_t prot)
1745{
Greg Ungerer66aa2b42005-09-12 11:18:10 +10001746 vma->vm_start = vma->vm_pgoff << PAGE_SHIFT;
1747 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748}
Luke Yang22c4af42006-07-14 00:24:09 -07001749EXPORT_SYMBOL(remap_pfn_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750
Paul Mundtf905bc42008-02-04 22:29:59 -08001751int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1752 unsigned long pgoff)
1753{
1754 unsigned int size = vma->vm_end - vma->vm_start;
1755
1756 if (!(vma->vm_flags & VM_USERMAP))
1757 return -EINVAL;
1758
1759 vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1760 vma->vm_end = vma->vm_start + size;
1761
1762 return 0;
1763}
1764EXPORT_SYMBOL(remap_vmalloc_range);
1765
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1767{
1768}
1769
1770unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1771 unsigned long len, unsigned long pgoff, unsigned long flags)
1772{
1773 return -ENOMEM;
1774}
1775
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001776void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777{
1778}
1779
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780void unmap_mapping_range(struct address_space *mapping,
1781 loff_t const holebegin, loff_t const holelen,
1782 int even_cows)
1783{
1784}
Luke Yang22c4af42006-07-14 00:24:09 -07001785EXPORT_SYMBOL(unmap_mapping_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786
1787/*
1788 * Check that a process has enough memory to allocate a new virtual
1789 * mapping. 0 means there is enough memory for the allocation to
1790 * succeed and -ENOMEM implies there is not.
1791 *
1792 * We currently support three overcommit policies, which are set via the
1793 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
1794 *
1795 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1796 * Additional code 2002 Jul 20 by Robert Love.
1797 *
1798 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1799 *
1800 * Note this is a helper function intended to be used by LSMs which
1801 * wish to use this logic.
1802 */
Alan Cox34b4e4a2007-08-22 14:01:28 -07001803int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804{
1805 unsigned long free, allowed;
1806
1807 vm_acct_memory(pages);
1808
1809 /*
1810 * Sometimes we want to use more memory than we have
1811 */
1812 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1813 return 0;
1814
1815 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1816 unsigned long n;
1817
Christoph Lameter347ce432006-06-30 01:55:35 -07001818 free = global_page_state(NR_FILE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819 free += nr_swap_pages;
1820
1821 /*
1822 * Any slabs which are created with the
1823 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1824 * which are reclaimable, under pressure. The dentry
1825 * cache and most inode caches should fall into this
1826 */
Christoph Lameter972d1a72006-09-25 23:31:51 -07001827 free += global_page_state(NR_SLAB_RECLAIMABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828
1829 /*
1830 * Leave the last 3% for root
1831 */
1832 if (!cap_sys_admin)
1833 free -= free / 32;
1834
1835 if (free > pages)
1836 return 0;
1837
1838 /*
1839 * nr_free_pages() is very expensive on large systems,
1840 * only call if we're about to fail.
1841 */
1842 n = nr_free_pages();
Hideo AOKId5ddc792006-04-10 22:53:01 -07001843
1844 /*
1845 * Leave reserved pages. The pages are not for anonymous pages.
1846 */
1847 if (n <= totalreserve_pages)
1848 goto error;
1849 else
1850 n -= totalreserve_pages;
1851
1852 /*
1853 * Leave the last 3% for root
1854 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855 if (!cap_sys_admin)
1856 n -= n / 32;
1857 free += n;
1858
1859 if (free > pages)
1860 return 0;
Hideo AOKId5ddc792006-04-10 22:53:01 -07001861
1862 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863 }
1864
1865 allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1866 /*
1867 * Leave the last 3% for root
1868 */
1869 if (!cap_sys_admin)
1870 allowed -= allowed / 32;
1871 allowed += total_swap_pages;
1872
1873 /* Don't let a single process grow too big:
1874 leave 3% of the size of this process for other processes */
Alan Cox731572d2008-10-29 14:01:20 -07001875 if (mm)
1876 allowed -= mm->total_vm / 32;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07001878 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879 return 0;
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07001880
Hideo AOKId5ddc792006-04-10 22:53:01 -07001881error:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882 vm_unacct_memory(pages);
1883
1884 return -ENOMEM;
1885}
1886
1887int in_gate_area_no_task(unsigned long addr)
1888{
1889 return 0;
1890}
David Howellsb0e15192006-01-06 00:11:42 -08001891
Nick Piggind0217ac2007-07-19 01:47:03 -07001892int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
David Howellsb0e15192006-01-06 00:11:42 -08001893{
1894 BUG();
Nick Piggind0217ac2007-07-19 01:47:03 -07001895 return 0;
David Howellsb0e15192006-01-06 00:11:42 -08001896}
Paul Mundtb5073172007-07-21 04:37:25 -07001897EXPORT_SYMBOL(filemap_fault);
David Howells0ec76a12006-09-27 01:50:15 -07001898
1899/*
1900 * Access another process' address space.
1901 * - source/target buffer must be kernel space
1902 */
1903int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
1904{
David Howells0ec76a12006-09-27 01:50:15 -07001905 struct vm_area_struct *vma;
1906 struct mm_struct *mm;
1907
1908 if (addr + len < addr)
1909 return 0;
1910
1911 mm = get_task_mm(tsk);
1912 if (!mm)
1913 return 0;
1914
1915 down_read(&mm->mmap_sem);
1916
1917 /* the access must start within one of the target process's mappings */
David Howells0159b142006-09-27 01:50:16 -07001918 vma = find_vma(mm, addr);
1919 if (vma) {
David Howells0ec76a12006-09-27 01:50:15 -07001920 /* don't overrun this mapping */
1921 if (addr + len >= vma->vm_end)
1922 len = vma->vm_end - addr;
1923
1924 /* only read or write mappings where it is permitted */
David Howellsd00c7b992006-09-27 01:50:19 -07001925 if (write && vma->vm_flags & VM_MAYWRITE)
Jie Zhang79597222010-01-06 17:23:28 +00001926 copy_to_user_page(vma, NULL, addr,
1927 (void *) addr, buf, len);
David Howellsd00c7b992006-09-27 01:50:19 -07001928 else if (!write && vma->vm_flags & VM_MAYREAD)
Jie Zhang79597222010-01-06 17:23:28 +00001929 copy_from_user_page(vma, NULL, addr,
1930 buf, (void *) addr, len);
David Howells0ec76a12006-09-27 01:50:15 -07001931 else
1932 len = 0;
1933 } else {
1934 len = 0;
1935 }
1936
1937 up_read(&mm->mmap_sem);
1938 mmput(mm);
1939 return len;
1940}
David Howells7e660872010-01-15 17:01:39 -08001941
1942/**
1943 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
1944 * @inode: The inode to check
1945 * @size: The current filesize of the inode
1946 * @newsize: The proposed filesize of the inode
1947 *
1948 * Check the shared mappings on an inode on behalf of a shrinking truncate to
1949 * make sure that that any outstanding VMAs aren't broken and then shrink the
1950 * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
1951 * automatically grant mappings that are too large.
1952 */
1953int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
1954 size_t newsize)
1955{
1956 struct vm_area_struct *vma;
1957 struct prio_tree_iter iter;
1958 struct vm_region *region;
1959 pgoff_t low, high;
1960 size_t r_size, r_top;
1961
1962 low = newsize >> PAGE_SHIFT;
1963 high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1964
1965 down_write(&nommu_region_sem);
1966
1967 /* search for VMAs that fall within the dead zone */
1968 vma_prio_tree_foreach(vma, &iter, &inode->i_mapping->i_mmap,
1969 low, high) {
1970 /* found one - only interested if it's shared out of the page
1971 * cache */
1972 if (vma->vm_flags & VM_SHARED) {
1973 up_write(&nommu_region_sem);
1974 return -ETXTBSY; /* not quite true, but near enough */
1975 }
1976 }
1977
1978 /* reduce any regions that overlap the dead zone - if in existence,
1979 * these will be pointed to by VMAs that don't overlap the dead zone
1980 *
1981 * we don't check for any regions that start beyond the EOF as there
1982 * shouldn't be any
1983 */
1984 vma_prio_tree_foreach(vma, &iter, &inode->i_mapping->i_mmap,
1985 0, ULONG_MAX) {
1986 if (!(vma->vm_flags & VM_SHARED))
1987 continue;
1988
1989 region = vma->vm_region;
1990 r_size = region->vm_top - region->vm_start;
1991 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
1992
1993 if (r_top > newsize) {
1994 region->vm_top -= r_top - newsize;
1995 if (region->vm_end > region->vm_top)
1996 region->vm_end = region->vm_top;
1997 }
1998 }
1999
2000 up_write(&nommu_region_sem);
2001 return 0;
2002}