blob: 8687973462bbea8becddfe0941eb94fab60ce304 [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++) {
Sonic Zhang910e46d2006-09-27 01:50:17 -0700149 vma = find_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 */
435 return mm->brk = brk;
436}
437
David Howells8feae132009-01-08 12:04:47 +0000438/*
439 * initialise the VMA and region record slabs
440 */
441void __init mmap_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442{
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -0700443 int ret;
444
445 ret = percpu_counter_init(&vm_committed_as, 0);
446 VM_BUG_ON(ret);
David Howells33e5d7692009-04-02 16:56:32 -0700447 vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC);
David Howells8feae132009-01-08 12:04:47 +0000448}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449
David Howells8feae132009-01-08 12:04:47 +0000450/*
451 * validate the region tree
452 * - the caller must hold the region lock
453 */
454#ifdef CONFIG_DEBUG_NOMMU_REGIONS
455static noinline void validate_nommu_regions(void)
456{
457 struct vm_region *region, *last;
458 struct rb_node *p, *lastp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459
David Howells8feae132009-01-08 12:04:47 +0000460 lastp = rb_first(&nommu_region_tree);
461 if (!lastp)
462 return;
463
464 last = rb_entry(lastp, struct vm_region, vm_rb);
David Howells33e5d7692009-04-02 16:56:32 -0700465 BUG_ON(unlikely(last->vm_end <= last->vm_start));
466 BUG_ON(unlikely(last->vm_top < last->vm_end));
David Howells8feae132009-01-08 12:04:47 +0000467
468 while ((p = rb_next(lastp))) {
469 region = rb_entry(p, struct vm_region, vm_rb);
470 last = rb_entry(lastp, struct vm_region, vm_rb);
471
David Howells33e5d7692009-04-02 16:56:32 -0700472 BUG_ON(unlikely(region->vm_end <= region->vm_start));
473 BUG_ON(unlikely(region->vm_top < region->vm_end));
474 BUG_ON(unlikely(region->vm_start < last->vm_top));
David Howells8feae132009-01-08 12:04:47 +0000475
476 lastp = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 }
478}
David Howells8feae132009-01-08 12:04:47 +0000479#else
David Howells33e5d7692009-04-02 16:56:32 -0700480static void validate_nommu_regions(void)
481{
482}
David Howells8feae132009-01-08 12:04:47 +0000483#endif
484
485/*
486 * add a region into the global tree
487 */
488static void add_nommu_region(struct vm_region *region)
489{
490 struct vm_region *pregion;
491 struct rb_node **p, *parent;
492
493 validate_nommu_regions();
494
David Howells8feae132009-01-08 12:04:47 +0000495 parent = NULL;
496 p = &nommu_region_tree.rb_node;
497 while (*p) {
498 parent = *p;
499 pregion = rb_entry(parent, struct vm_region, vm_rb);
500 if (region->vm_start < pregion->vm_start)
501 p = &(*p)->rb_left;
502 else if (region->vm_start > pregion->vm_start)
503 p = &(*p)->rb_right;
504 else if (pregion == region)
505 return;
506 else
507 BUG();
508 }
509
510 rb_link_node(&region->vm_rb, parent, p);
511 rb_insert_color(&region->vm_rb, &nommu_region_tree);
512
513 validate_nommu_regions();
514}
515
516/*
517 * delete a region from the global tree
518 */
519static void delete_nommu_region(struct vm_region *region)
520{
521 BUG_ON(!nommu_region_tree.rb_node);
522
523 validate_nommu_regions();
524 rb_erase(&region->vm_rb, &nommu_region_tree);
525 validate_nommu_regions();
526}
527
528/*
529 * free a contiguous series of pages
530 */
531static void free_page_series(unsigned long from, unsigned long to)
532{
533 for (; from < to; from += PAGE_SIZE) {
534 struct page *page = virt_to_page(from);
535
536 kdebug("- free %lx", from);
David Howells33e5d7692009-04-02 16:56:32 -0700537 atomic_long_dec(&mmap_pages_allocated);
David Howells8feae132009-01-08 12:04:47 +0000538 if (page_count(page) != 1)
David Howells33e5d7692009-04-02 16:56:32 -0700539 kdebug("free page %p: refcount not one: %d",
540 page, page_count(page));
David Howells8feae132009-01-08 12:04:47 +0000541 put_page(page);
542 }
543}
544
545/*
546 * release a reference to a region
David Howells33e5d7692009-04-02 16:56:32 -0700547 * - the caller must hold the region semaphore for writing, which this releases
Paul Mundtdd8632a2009-01-08 12:04:47 +0000548 * - the region may not have been added to the tree yet, in which case vm_top
David Howells8feae132009-01-08 12:04:47 +0000549 * will equal vm_start
550 */
551static void __put_nommu_region(struct vm_region *region)
552 __releases(nommu_region_sem)
553{
554 kenter("%p{%d}", region, atomic_read(&region->vm_usage));
555
556 BUG_ON(!nommu_region_tree.rb_node);
557
558 if (atomic_dec_and_test(&region->vm_usage)) {
Paul Mundtdd8632a2009-01-08 12:04:47 +0000559 if (region->vm_top > region->vm_start)
David Howells8feae132009-01-08 12:04:47 +0000560 delete_nommu_region(region);
561 up_write(&nommu_region_sem);
562
563 if (region->vm_file)
564 fput(region->vm_file);
565
566 /* IO memory and memory shared directly out of the pagecache
567 * from ramfs/tmpfs mustn't be released here */
568 if (region->vm_flags & VM_MAPPED_COPY) {
569 kdebug("free series");
Paul Mundtdd8632a2009-01-08 12:04:47 +0000570 free_page_series(region->vm_start, region->vm_top);
David Howells8feae132009-01-08 12:04:47 +0000571 }
572 kmem_cache_free(vm_region_jar, region);
573 } else {
574 up_write(&nommu_region_sem);
575 }
576}
577
578/*
579 * release a reference to a region
580 */
581static void put_nommu_region(struct vm_region *region)
582{
583 down_write(&nommu_region_sem);
584 __put_nommu_region(region);
585}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586
David Howells30340972006-09-27 01:50:20 -0700587/*
Bernd Schmidteb8cdec2009-09-21 17:03:57 -0700588 * update protection on a vma
589 */
590static void protect_vma(struct vm_area_struct *vma, unsigned long flags)
591{
592#ifdef CONFIG_MPU
593 struct mm_struct *mm = vma->vm_mm;
594 long start = vma->vm_start & PAGE_MASK;
595 while (start < vma->vm_end) {
596 protect_page(mm, start, flags);
597 start += PAGE_SIZE;
598 }
599 update_protections(mm);
600#endif
601}
602
603/*
David Howells30340972006-09-27 01:50:20 -0700604 * add a VMA into a process's mm_struct in the appropriate place in the list
David Howells8feae132009-01-08 12:04:47 +0000605 * and tree and add to the address space's page tree also if not an anonymous
606 * page
David Howells30340972006-09-27 01:50:20 -0700607 * - should be called with mm->mmap_sem held writelocked
608 */
David Howells8feae132009-01-08 12:04:47 +0000609static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
David Howells30340972006-09-27 01:50:20 -0700610{
David Howells8feae132009-01-08 12:04:47 +0000611 struct vm_area_struct *pvma, **pp;
612 struct address_space *mapping;
613 struct rb_node **p, *parent;
David Howells30340972006-09-27 01:50:20 -0700614
David Howells8feae132009-01-08 12:04:47 +0000615 kenter(",%p", vma);
616
617 BUG_ON(!vma->vm_region);
618
619 mm->map_count++;
620 vma->vm_mm = mm;
621
Bernd Schmidteb8cdec2009-09-21 17:03:57 -0700622 protect_vma(vma, vma->vm_flags);
623
David Howells8feae132009-01-08 12:04:47 +0000624 /* add the VMA to the mapping */
625 if (vma->vm_file) {
626 mapping = vma->vm_file->f_mapping;
627
628 flush_dcache_mmap_lock(mapping);
629 vma_prio_tree_insert(vma, &mapping->i_mmap);
630 flush_dcache_mmap_unlock(mapping);
631 }
632
633 /* add the VMA to the tree */
634 parent = NULL;
635 p = &mm->mm_rb.rb_node;
636 while (*p) {
637 parent = *p;
638 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
639
640 /* sort by: start addr, end addr, VMA struct addr in that order
641 * (the latter is necessary as we may get identical VMAs) */
642 if (vma->vm_start < pvma->vm_start)
643 p = &(*p)->rb_left;
644 else if (vma->vm_start > pvma->vm_start)
645 p = &(*p)->rb_right;
646 else if (vma->vm_end < pvma->vm_end)
647 p = &(*p)->rb_left;
648 else if (vma->vm_end > pvma->vm_end)
649 p = &(*p)->rb_right;
650 else if (vma < pvma)
651 p = &(*p)->rb_left;
652 else if (vma > pvma)
653 p = &(*p)->rb_right;
654 else
655 BUG();
656 }
657
658 rb_link_node(&vma->vm_rb, parent, p);
659 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
660
661 /* add VMA to the VMA list also */
662 for (pp = &mm->mmap; (pvma = *pp); pp = &(*pp)->vm_next) {
663 if (pvma->vm_start > vma->vm_start)
David Howells30340972006-09-27 01:50:20 -0700664 break;
David Howells8feae132009-01-08 12:04:47 +0000665 if (pvma->vm_start < vma->vm_start)
666 continue;
667 if (pvma->vm_end < vma->vm_end)
668 break;
669 }
David Howells30340972006-09-27 01:50:20 -0700670
David Howells8feae132009-01-08 12:04:47 +0000671 vma->vm_next = *pp;
672 *pp = vma;
673}
674
675/*
676 * delete a VMA from its owning mm_struct and address space
677 */
678static void delete_vma_from_mm(struct vm_area_struct *vma)
679{
680 struct vm_area_struct **pp;
681 struct address_space *mapping;
682 struct mm_struct *mm = vma->vm_mm;
683
684 kenter("%p", vma);
685
Bernd Schmidteb8cdec2009-09-21 17:03:57 -0700686 protect_vma(vma, 0);
687
David Howells8feae132009-01-08 12:04:47 +0000688 mm->map_count--;
689 if (mm->mmap_cache == vma)
690 mm->mmap_cache = NULL;
691
692 /* remove the VMA from the mapping */
693 if (vma->vm_file) {
694 mapping = vma->vm_file->f_mapping;
695
696 flush_dcache_mmap_lock(mapping);
697 vma_prio_tree_remove(vma, &mapping->i_mmap);
698 flush_dcache_mmap_unlock(mapping);
699 }
700
701 /* remove from the MM's tree and list */
702 rb_erase(&vma->vm_rb, &mm->mm_rb);
703 for (pp = &mm->mmap; *pp; pp = &(*pp)->vm_next) {
704 if (*pp == vma) {
705 *pp = vma->vm_next;
706 break;
707 }
708 }
709
710 vma->vm_mm = NULL;
711}
712
713/*
714 * destroy a VMA record
715 */
716static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
717{
718 kenter("%p", vma);
719 if (vma->vm_ops && vma->vm_ops->close)
720 vma->vm_ops->close(vma);
721 if (vma->vm_file) {
722 fput(vma->vm_file);
723 if (vma->vm_flags & VM_EXECUTABLE)
724 removed_exe_file_vma(mm);
725 }
726 put_nommu_region(vma->vm_region);
727 kmem_cache_free(vm_area_cachep, vma);
David Howells30340972006-09-27 01:50:20 -0700728}
729
730/*
731 * look up the first VMA in which addr resides, NULL if none
732 * - should be called with mm->mmap_sem at least held readlocked
733 */
734struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
735{
David Howells8feae132009-01-08 12:04:47 +0000736 struct vm_area_struct *vma;
737 struct rb_node *n = mm->mm_rb.rb_node;
David Howells30340972006-09-27 01:50:20 -0700738
David Howells8feae132009-01-08 12:04:47 +0000739 /* check the cache first */
740 vma = mm->mmap_cache;
741 if (vma && vma->vm_start <= addr && vma->vm_end > addr)
742 return vma;
743
744 /* trawl the tree (there may be multiple mappings in which addr
745 * resides) */
746 for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) {
747 vma = rb_entry(n, struct vm_area_struct, vm_rb);
748 if (vma->vm_start > addr)
749 return NULL;
750 if (vma->vm_end > addr) {
751 mm->mmap_cache = vma;
752 return vma;
753 }
David Howells30340972006-09-27 01:50:20 -0700754 }
755
David Howells30340972006-09-27 01:50:20 -0700756 return NULL;
757}
758EXPORT_SYMBOL(find_vma);
759
760/*
David Howells930e6522006-09-27 01:50:22 -0700761 * find a VMA
762 * - we don't extend stack VMAs under NOMMU conditions
763 */
764struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
765{
766 return find_vma(mm, addr);
767}
768
David Howells8feae132009-01-08 12:04:47 +0000769/*
770 * expand a stack to a given address
771 * - not supported under NOMMU conditions
772 */
Greg Ungerer57c8f632007-07-15 23:38:28 -0700773int expand_stack(struct vm_area_struct *vma, unsigned long address)
774{
775 return -ENOMEM;
776}
777
David Howells930e6522006-09-27 01:50:22 -0700778/*
David Howells6fa5f802006-09-27 01:50:21 -0700779 * look up the first VMA exactly that exactly matches addr
780 * - should be called with mm->mmap_sem at least held readlocked
781 */
David Howells8feae132009-01-08 12:04:47 +0000782static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
783 unsigned long addr,
784 unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785{
786 struct vm_area_struct *vma;
David Howells8feae132009-01-08 12:04:47 +0000787 struct rb_node *n = mm->mm_rb.rb_node;
788 unsigned long end = addr + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789
David Howells8feae132009-01-08 12:04:47 +0000790 /* check the cache first */
791 vma = mm->mmap_cache;
792 if (vma && vma->vm_start == addr && vma->vm_end == end)
793 return vma;
794
795 /* trawl the tree (there may be multiple mappings in which addr
796 * resides) */
797 for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 vma = rb_entry(n, struct vm_area_struct, vm_rb);
David Howells8feae132009-01-08 12:04:47 +0000799 if (vma->vm_start < addr)
800 continue;
801 if (vma->vm_start > addr)
802 return NULL;
803 if (vma->vm_end == end) {
804 mm->mmap_cache = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 return vma;
David Howells8feae132009-01-08 12:04:47 +0000806 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 }
808
809 return NULL;
810}
811
David Howells30340972006-09-27 01:50:20 -0700812/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 * determine whether a mapping should be permitted and, if so, what sort of
814 * mapping we're capable of supporting
815 */
816static int validate_mmap_request(struct file *file,
817 unsigned long addr,
818 unsigned long len,
819 unsigned long prot,
820 unsigned long flags,
821 unsigned long pgoff,
822 unsigned long *_capabilities)
823{
David Howells8feae132009-01-08 12:04:47 +0000824 unsigned long capabilities, rlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 unsigned long reqprot = prot;
826 int ret;
827
828 /* do the simple checks first */
David Howells06aab5a2009-09-24 12:33:48 +0100829 if (flags & MAP_FIXED) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 printk(KERN_DEBUG
831 "%d: Can't do fixed-address/overlay mmap of RAM\n",
832 current->pid);
833 return -EINVAL;
834 }
835
836 if ((flags & MAP_TYPE) != MAP_PRIVATE &&
837 (flags & MAP_TYPE) != MAP_SHARED)
838 return -EINVAL;
839
Mike Frysingerf81cff02006-12-06 12:02:59 +1000840 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 return -EINVAL;
842
Mike Frysingerf81cff02006-12-06 12:02:59 +1000843 /* Careful about overflows.. */
David Howells8feae132009-01-08 12:04:47 +0000844 rlen = PAGE_ALIGN(len);
845 if (!rlen || rlen > TASK_SIZE)
Mike Frysingerf81cff02006-12-06 12:02:59 +1000846 return -ENOMEM;
847
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848 /* offset overflow? */
David Howells8feae132009-01-08 12:04:47 +0000849 if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
Mike Frysingerf81cff02006-12-06 12:02:59 +1000850 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851
852 if (file) {
853 /* validate file mapping requests */
854 struct address_space *mapping;
855
856 /* files must support mmap */
857 if (!file->f_op || !file->f_op->mmap)
858 return -ENODEV;
859
860 /* work out if what we've got could possibly be shared
861 * - we support chardevs that provide their own "memory"
862 * - we support files/blockdevs that are memory backed
863 */
864 mapping = file->f_mapping;
865 if (!mapping)
Josef Sipeke9536ae2006-12-08 02:37:21 -0800866 mapping = file->f_path.dentry->d_inode->i_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867
868 capabilities = 0;
869 if (mapping && mapping->backing_dev_info)
870 capabilities = mapping->backing_dev_info->capabilities;
871
872 if (!capabilities) {
873 /* no explicit capabilities set, so assume some
874 * defaults */
Josef Sipeke9536ae2006-12-08 02:37:21 -0800875 switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 case S_IFREG:
877 case S_IFBLK:
878 capabilities = BDI_CAP_MAP_COPY;
879 break;
880
881 case S_IFCHR:
882 capabilities =
883 BDI_CAP_MAP_DIRECT |
884 BDI_CAP_READ_MAP |
885 BDI_CAP_WRITE_MAP;
886 break;
887
888 default:
889 return -EINVAL;
890 }
891 }
892
893 /* eliminate any capabilities that we can't support on this
894 * device */
895 if (!file->f_op->get_unmapped_area)
896 capabilities &= ~BDI_CAP_MAP_DIRECT;
897 if (!file->f_op->read)
898 capabilities &= ~BDI_CAP_MAP_COPY;
899
Graff Yang28d7a6a2009-08-18 14:11:17 -0700900 /* The file shall have been opened with read permission. */
901 if (!(file->f_mode & FMODE_READ))
902 return -EACCES;
903
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904 if (flags & MAP_SHARED) {
905 /* do checks for writing, appending and locking */
906 if ((prot & PROT_WRITE) &&
907 !(file->f_mode & FMODE_WRITE))
908 return -EACCES;
909
Josef Sipeke9536ae2006-12-08 02:37:21 -0800910 if (IS_APPEND(file->f_path.dentry->d_inode) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 (file->f_mode & FMODE_WRITE))
912 return -EACCES;
913
Josef Sipeke9536ae2006-12-08 02:37:21 -0800914 if (locks_verify_locked(file->f_path.dentry->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 return -EAGAIN;
916
917 if (!(capabilities & BDI_CAP_MAP_DIRECT))
918 return -ENODEV;
919
920 if (((prot & PROT_READ) && !(capabilities & BDI_CAP_READ_MAP)) ||
921 ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
922 ((prot & PROT_EXEC) && !(capabilities & BDI_CAP_EXEC_MAP))
923 ) {
924 printk("MAP_SHARED not completely supported on !MMU\n");
925 return -EINVAL;
926 }
927
928 /* we mustn't privatise shared mappings */
929 capabilities &= ~BDI_CAP_MAP_COPY;
930 }
931 else {
932 /* we're going to read the file into private memory we
933 * allocate */
934 if (!(capabilities & BDI_CAP_MAP_COPY))
935 return -ENODEV;
936
937 /* we don't permit a private writable mapping to be
938 * shared with the backing device */
939 if (prot & PROT_WRITE)
940 capabilities &= ~BDI_CAP_MAP_DIRECT;
941 }
942
943 /* handle executable mappings and implied executable
944 * mappings */
Josef Sipeke9536ae2006-12-08 02:37:21 -0800945 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 if (prot & PROT_EXEC)
947 return -EPERM;
948 }
949 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
950 /* handle implication of PROT_EXEC by PROT_READ */
951 if (current->personality & READ_IMPLIES_EXEC) {
952 if (capabilities & BDI_CAP_EXEC_MAP)
953 prot |= PROT_EXEC;
954 }
955 }
956 else if ((prot & PROT_READ) &&
957 (prot & PROT_EXEC) &&
958 !(capabilities & BDI_CAP_EXEC_MAP)
959 ) {
960 /* backing file is not executable, try to copy */
961 capabilities &= ~BDI_CAP_MAP_DIRECT;
962 }
963 }
964 else {
965 /* anonymous mappings are always memory backed and can be
966 * privately mapped
967 */
968 capabilities = BDI_CAP_MAP_COPY;
969
970 /* handle PROT_EXEC implication by PROT_READ */
971 if ((prot & PROT_READ) &&
972 (current->personality & READ_IMPLIES_EXEC))
973 prot |= PROT_EXEC;
974 }
975
976 /* allow the security API to have its say */
Eric Parised032182007-06-28 15:55:21 -0400977 ret = security_file_mmap(file, reqprot, prot, flags, addr, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 if (ret < 0)
979 return ret;
980
981 /* looks okay */
982 *_capabilities = capabilities;
983 return 0;
984}
985
986/*
987 * we've determined that we can make the mapping, now translate what we
988 * now know into VMA flags
989 */
990static unsigned long determine_vm_flags(struct file *file,
991 unsigned long prot,
992 unsigned long flags,
993 unsigned long capabilities)
994{
995 unsigned long vm_flags;
996
997 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
998 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
999 /* vm_flags |= mm->def_flags; */
1000
1001 if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
1002 /* attempt to share read-only copies of mapped file chunks */
1003 if (file && !(prot & PROT_WRITE))
1004 vm_flags |= VM_MAYSHARE;
1005 }
1006 else {
1007 /* overlay a shareable mapping on the backing device or inode
1008 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1009 * romfs/cramfs */
1010 if (flags & MAP_SHARED)
1011 vm_flags |= VM_MAYSHARE | VM_SHARED;
1012 else if ((((vm_flags & capabilities) ^ vm_flags) & BDI_CAP_VMFLAGS) == 0)
1013 vm_flags |= VM_MAYSHARE;
1014 }
1015
1016 /* refuse to let anyone share private mappings with this process if
1017 * it's being traced - otherwise breakpoints set in it may interfere
1018 * with another untraced process
1019 */
Roland McGrathfa8e26c2008-07-25 19:45:50 -07001020 if ((flags & MAP_PRIVATE) && tracehook_expect_breakpoints(current))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 vm_flags &= ~VM_MAYSHARE;
1022
1023 return vm_flags;
1024}
1025
1026/*
David Howells8feae132009-01-08 12:04:47 +00001027 * set up a shared mapping on a file (the driver or filesystem provides and
1028 * pins the storage)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 */
David Howells8feae132009-01-08 12:04:47 +00001030static int do_mmap_shared_file(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031{
1032 int ret;
1033
1034 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001035 if (ret == 0) {
1036 vma->vm_region->vm_top = vma->vm_region->vm_end;
David Howells645d83c2009-09-24 15:13:10 +01001037 return 0;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001038 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 if (ret != -ENOSYS)
1040 return ret;
1041
1042 /* getting an ENOSYS error indicates that direct mmap isn't
1043 * possible (as opposed to tried but failed) so we'll fall
1044 * through to making a private copy of the data and mapping
1045 * that if we can */
1046 return -ENODEV;
1047}
1048
1049/*
1050 * set up a private mapping or an anonymous shared mapping
1051 */
David Howells8feae132009-01-08 12:04:47 +00001052static int do_mmap_private(struct vm_area_struct *vma,
1053 struct vm_region *region,
David Howells645d83c2009-09-24 15:13:10 +01001054 unsigned long len,
1055 unsigned long capabilities)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056{
David Howells8feae132009-01-08 12:04:47 +00001057 struct page *pages;
1058 unsigned long total, point, n, rlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 void *base;
David Howells8feae132009-01-08 12:04:47 +00001060 int ret, order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061
1062 /* invoke the file's mapping function so that it can keep track of
1063 * shared mappings on devices or memory
1064 * - VM_MAYSHARE will be set if it may attempt to share
1065 */
David Howells645d83c2009-09-24 15:13:10 +01001066 if (capabilities & BDI_CAP_MAP_DIRECT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001068 if (ret == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069 /* shouldn't return success if we're not sharing */
Paul Mundtdd8632a2009-01-08 12:04:47 +00001070 BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1071 vma->vm_region->vm_top = vma->vm_region->vm_end;
David Howells645d83c2009-09-24 15:13:10 +01001072 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 }
Paul Mundtdd8632a2009-01-08 12:04:47 +00001074 if (ret != -ENOSYS)
1075 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076
1077 /* getting an ENOSYS error indicates that direct mmap isn't
1078 * possible (as opposed to tried but failed) so we'll try to
1079 * make a private copy of the data and map that instead */
1080 }
1081
David Howells8feae132009-01-08 12:04:47 +00001082 rlen = PAGE_ALIGN(len);
1083
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 /* allocate some memory to hold the mapping
1085 * - note that this may not return a page-aligned address if the object
1086 * we're allocating is smaller than a page
1087 */
David Howells8feae132009-01-08 12:04:47 +00001088 order = get_order(rlen);
1089 kdebug("alloc order %d for %lx", order, len);
1090
1091 pages = alloc_pages(GFP_KERNEL, order);
1092 if (!pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 goto enomem;
1094
David Howells8feae132009-01-08 12:04:47 +00001095 total = 1 << order;
David Howells33e5d7692009-04-02 16:56:32 -07001096 atomic_long_add(total, &mmap_pages_allocated);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097
David Howells8feae132009-01-08 12:04:47 +00001098 point = rlen >> PAGE_SHIFT;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001099
1100 /* we allocated a power-of-2 sized page set, so we may want to trim off
1101 * the excess */
1102 if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) {
1103 while (total > point) {
1104 order = ilog2(total - point);
1105 n = 1 << order;
1106 kdebug("shave %lu/%lu @%lu", n, total - point, total);
David Howells33e5d7692009-04-02 16:56:32 -07001107 atomic_long_sub(n, &mmap_pages_allocated);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001108 total -= n;
1109 set_page_refcounted(pages + total);
1110 __free_pages(pages + total, order);
1111 }
David Howells8feae132009-01-08 12:04:47 +00001112 }
1113
David Howells8feae132009-01-08 12:04:47 +00001114 for (point = 1; point < total; point++)
1115 set_page_refcounted(&pages[point]);
1116
1117 base = page_address(pages);
1118 region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1119 region->vm_start = (unsigned long) base;
1120 region->vm_end = region->vm_start + rlen;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001121 region->vm_top = region->vm_start + (total << PAGE_SHIFT);
David Howells8feae132009-01-08 12:04:47 +00001122
1123 vma->vm_start = region->vm_start;
1124 vma->vm_end = region->vm_start + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125
1126 if (vma->vm_file) {
1127 /* read the contents of a file into the copy */
1128 mm_segment_t old_fs;
1129 loff_t fpos;
1130
1131 fpos = vma->vm_pgoff;
1132 fpos <<= PAGE_SHIFT;
1133
1134 old_fs = get_fs();
1135 set_fs(KERNEL_DS);
David Howells8feae132009-01-08 12:04:47 +00001136 ret = vma->vm_file->f_op->read(vma->vm_file, base, rlen, &fpos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137 set_fs(old_fs);
1138
1139 if (ret < 0)
1140 goto error_free;
1141
1142 /* clear the last little bit */
David Howells8feae132009-01-08 12:04:47 +00001143 if (ret < rlen)
1144 memset(base + ret, 0, rlen - ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 }
1147
1148 return 0;
1149
1150error_free:
David Howells8feae132009-01-08 12:04:47 +00001151 free_page_series(region->vm_start, region->vm_end);
1152 region->vm_start = vma->vm_start = 0;
1153 region->vm_end = vma->vm_end = 0;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001154 region->vm_top = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155 return ret;
1156
1157enomem:
Greg Ungerer05ae6fa2009-01-13 17:30:22 +10001158 printk("Allocation of length %lu from process %d (%s) failed\n",
1159 len, current->pid, current->comm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 show_free_areas();
1161 return -ENOMEM;
1162}
1163
1164/*
1165 * handle mapping creation for uClinux
1166 */
1167unsigned long do_mmap_pgoff(struct file *file,
1168 unsigned long addr,
1169 unsigned long len,
1170 unsigned long prot,
1171 unsigned long flags,
1172 unsigned long pgoff)
1173{
David Howells8feae132009-01-08 12:04:47 +00001174 struct vm_area_struct *vma;
1175 struct vm_region *region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176 struct rb_node *rb;
David Howells8feae132009-01-08 12:04:47 +00001177 unsigned long capabilities, vm_flags, result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 int ret;
1179
David Howells8feae132009-01-08 12:04:47 +00001180 kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff);
1181
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 /* decide whether we should attempt the mapping, and if so what sort of
1183 * mapping */
1184 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1185 &capabilities);
David Howells8feae132009-01-08 12:04:47 +00001186 if (ret < 0) {
1187 kleave(" = %d [val]", ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 return ret;
David Howells8feae132009-01-08 12:04:47 +00001189 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190
David Howells06aab5a2009-09-24 12:33:48 +01001191 /* we ignore the address hint */
1192 addr = 0;
1193
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 /* we've determined that we can make the mapping, now translate what we
1195 * now know into VMA flags */
1196 vm_flags = determine_vm_flags(file, prot, flags, capabilities);
1197
David Howells8feae132009-01-08 12:04:47 +00001198 /* we're going to need to record the mapping */
1199 region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1200 if (!region)
1201 goto error_getting_region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202
David Howells8feae132009-01-08 12:04:47 +00001203 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1204 if (!vma)
1205 goto error_getting_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206
David Howells8feae132009-01-08 12:04:47 +00001207 atomic_set(&region->vm_usage, 1);
1208 region->vm_flags = vm_flags;
1209 region->vm_pgoff = pgoff;
1210
1211 INIT_LIST_HEAD(&vma->anon_vma_node);
1212 vma->vm_flags = vm_flags;
1213 vma->vm_pgoff = pgoff;
1214
1215 if (file) {
1216 region->vm_file = file;
1217 get_file(file);
1218 vma->vm_file = file;
1219 get_file(file);
1220 if (vm_flags & VM_EXECUTABLE) {
1221 added_exe_file_vma(current->mm);
1222 vma->vm_mm = current->mm;
1223 }
1224 }
1225
1226 down_write(&nommu_region_sem);
1227
1228 /* if we want to share, we need to check for regions created by other
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 * mmap() calls that overlap with our proposed mapping
David Howells8feae132009-01-08 12:04:47 +00001230 * - we can only share with a superset match on most regular files
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 * - shared mappings on character devices and memory backed files are
1232 * permitted to overlap inexactly as far as we are concerned for in
1233 * these cases, sharing is handled in the driver or filesystem rather
1234 * than here
1235 */
1236 if (vm_flags & VM_MAYSHARE) {
David Howells8feae132009-01-08 12:04:47 +00001237 struct vm_region *pregion;
1238 unsigned long pglen, rpglen, pgend, rpgend, start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239
David Howells8feae132009-01-08 12:04:47 +00001240 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1241 pgend = pgoff + pglen;
David Howells165b2392007-03-22 00:11:24 -08001242
David Howells8feae132009-01-08 12:04:47 +00001243 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1244 pregion = rb_entry(rb, struct vm_region, vm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245
David Howells8feae132009-01-08 12:04:47 +00001246 if (!(pregion->vm_flags & VM_MAYSHARE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247 continue;
1248
1249 /* search for overlapping mappings on the same file */
David Howells8feae132009-01-08 12:04:47 +00001250 if (pregion->vm_file->f_path.dentry->d_inode !=
1251 file->f_path.dentry->d_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 continue;
1253
David Howells8feae132009-01-08 12:04:47 +00001254 if (pregion->vm_pgoff >= pgend)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 continue;
1256
David Howells8feae132009-01-08 12:04:47 +00001257 rpglen = pregion->vm_end - pregion->vm_start;
1258 rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1259 rpgend = pregion->vm_pgoff + rpglen;
1260 if (pgoff >= rpgend)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 continue;
1262
David Howells8feae132009-01-08 12:04:47 +00001263 /* handle inexactly overlapping matches between
1264 * mappings */
1265 if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1266 !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1267 /* new mapping is not a subset of the region */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 if (!(capabilities & BDI_CAP_MAP_DIRECT))
1269 goto sharing_violation;
1270 continue;
1271 }
1272
David Howells8feae132009-01-08 12:04:47 +00001273 /* we've found a region we can share */
1274 atomic_inc(&pregion->vm_usage);
1275 vma->vm_region = pregion;
1276 start = pregion->vm_start;
1277 start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1278 vma->vm_start = start;
1279 vma->vm_end = start + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280
David Howells8feae132009-01-08 12:04:47 +00001281 if (pregion->vm_flags & VM_MAPPED_COPY) {
1282 kdebug("share copy");
1283 vma->vm_flags |= VM_MAPPED_COPY;
1284 } else {
1285 kdebug("share mmap");
1286 ret = do_mmap_shared_file(vma);
1287 if (ret < 0) {
1288 vma->vm_region = NULL;
1289 vma->vm_start = 0;
1290 vma->vm_end = 0;
1291 atomic_dec(&pregion->vm_usage);
1292 pregion = NULL;
1293 goto error_just_free;
1294 }
1295 }
1296 fput(region->vm_file);
1297 kmem_cache_free(vm_region_jar, region);
1298 region = pregion;
1299 result = start;
1300 goto share;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301 }
1302
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 /* obtain the address at which to make a shared mapping
1304 * - this is the hook for quasi-memory character devices to
1305 * tell us the location of a shared mapping
1306 */
David Howells645d83c2009-09-24 15:13:10 +01001307 if (capabilities & BDI_CAP_MAP_DIRECT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308 addr = file->f_op->get_unmapped_area(file, addr, len,
1309 pgoff, flags);
1310 if (IS_ERR((void *) addr)) {
1311 ret = addr;
1312 if (ret != (unsigned long) -ENOSYS)
David Howells8feae132009-01-08 12:04:47 +00001313 goto error_just_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314
1315 /* the driver refused to tell us where to site
1316 * the mapping so we'll have to attempt to copy
1317 * it */
1318 ret = (unsigned long) -ENODEV;
1319 if (!(capabilities & BDI_CAP_MAP_COPY))
David Howells8feae132009-01-08 12:04:47 +00001320 goto error_just_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321
1322 capabilities &= ~BDI_CAP_MAP_DIRECT;
David Howells8feae132009-01-08 12:04:47 +00001323 } else {
1324 vma->vm_start = region->vm_start = addr;
1325 vma->vm_end = region->vm_end = addr + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326 }
1327 }
1328 }
1329
David Howells8feae132009-01-08 12:04:47 +00001330 vma->vm_region = region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331
David Howells645d83c2009-09-24 15:13:10 +01001332 /* set up the mapping
1333 * - the region is filled in if BDI_CAP_MAP_DIRECT is still set
1334 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 if (file && vma->vm_flags & VM_SHARED)
David Howells8feae132009-01-08 12:04:47 +00001336 ret = do_mmap_shared_file(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 else
David Howells645d83c2009-09-24 15:13:10 +01001338 ret = do_mmap_private(vma, region, len, capabilities);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 if (ret < 0)
David Howells645d83c2009-09-24 15:13:10 +01001340 goto error_just_free;
1341 add_nommu_region(region);
David Howells8feae132009-01-08 12:04:47 +00001342
Jie Zhangea637632009-12-14 18:00:02 -08001343 /* clear anonymous mappings that don't ask for uninitialized data */
1344 if (!vma->vm_file && !(flags & MAP_UNINITIALIZED))
1345 memset((void *)region->vm_start, 0,
1346 region->vm_end - region->vm_start);
1347
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 /* okay... we have a mapping; now we have to register it */
David Howells8feae132009-01-08 12:04:47 +00001349 result = vma->vm_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 current->mm->total_vm += len >> PAGE_SHIFT;
1352
David Howells8feae132009-01-08 12:04:47 +00001353share:
1354 add_vma_to_mm(current->mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355
David Howells8feae132009-01-08 12:04:47 +00001356 up_write(&nommu_region_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357
1358 if (prot & PROT_EXEC)
David Howells8feae132009-01-08 12:04:47 +00001359 flush_icache_range(result, result + len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360
David Howells8feae132009-01-08 12:04:47 +00001361 kleave(" = %lx", result);
1362 return result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363
David Howells8feae132009-01-08 12:04:47 +00001364error_just_free:
1365 up_write(&nommu_region_sem);
1366error:
David Howells89a86402009-10-30 13:13:26 +00001367 if (region->vm_file)
1368 fput(region->vm_file);
David Howells8feae132009-01-08 12:04:47 +00001369 kmem_cache_free(vm_region_jar, region);
David Howells89a86402009-10-30 13:13:26 +00001370 if (vma->vm_file)
1371 fput(vma->vm_file);
David Howells8feae132009-01-08 12:04:47 +00001372 if (vma->vm_flags & VM_EXECUTABLE)
1373 removed_exe_file_vma(vma->vm_mm);
1374 kmem_cache_free(vm_area_cachep, vma);
1375 kleave(" = %d", ret);
1376 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377
David Howells8feae132009-01-08 12:04:47 +00001378sharing_violation:
1379 up_write(&nommu_region_sem);
1380 printk(KERN_WARNING "Attempt to share mismatched mappings\n");
1381 ret = -EINVAL;
1382 goto error;
1383
1384error_getting_vma:
1385 kmem_cache_free(vm_region_jar, region);
1386 printk(KERN_WARNING "Allocation of vma for %lu byte allocation"
1387 " from process %d failed\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 len, current->pid);
1389 show_free_areas();
1390 return -ENOMEM;
1391
David Howells8feae132009-01-08 12:04:47 +00001392error_getting_region:
1393 printk(KERN_WARNING "Allocation of vm region for %lu byte allocation"
1394 " from process %d failed\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 len, current->pid);
1396 show_free_areas();
1397 return -ENOMEM;
1398}
Paul Mundtb5073172007-07-21 04:37:25 -07001399EXPORT_SYMBOL(do_mmap_pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400
1401/*
David Howells8feae132009-01-08 12:04:47 +00001402 * split a vma into two pieces at address 'addr', a new vma is allocated either
1403 * for the first part or the tail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 */
David Howells8feae132009-01-08 12:04:47 +00001405int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1406 unsigned long addr, int new_below)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407{
David Howells8feae132009-01-08 12:04:47 +00001408 struct vm_area_struct *new;
1409 struct vm_region *region;
1410 unsigned long npages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411
David Howells8feae132009-01-08 12:04:47 +00001412 kenter("");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413
David Howells8feae132009-01-08 12:04:47 +00001414 /* we're only permitted to split anonymous regions that have a single
1415 * owner */
1416 if (vma->vm_file ||
1417 atomic_read(&vma->vm_region->vm_usage) != 1)
1418 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419
David Howells8feae132009-01-08 12:04:47 +00001420 if (mm->map_count >= sysctl_max_map_count)
1421 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422
David Howells8feae132009-01-08 12:04:47 +00001423 region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1424 if (!region)
1425 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426
David Howells8feae132009-01-08 12:04:47 +00001427 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1428 if (!new) {
1429 kmem_cache_free(vm_region_jar, region);
1430 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431 }
David Howells8feae132009-01-08 12:04:47 +00001432
1433 /* most fields are the same, copy all, and then fixup */
1434 *new = *vma;
1435 *region = *vma->vm_region;
1436 new->vm_region = region;
1437
1438 npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1439
1440 if (new_below) {
Paul Mundtdd8632a2009-01-08 12:04:47 +00001441 region->vm_top = region->vm_end = new->vm_end = addr;
David Howells8feae132009-01-08 12:04:47 +00001442 } else {
1443 region->vm_start = new->vm_start = addr;
1444 region->vm_pgoff = new->vm_pgoff += npages;
1445 }
1446
1447 if (new->vm_ops && new->vm_ops->open)
1448 new->vm_ops->open(new);
1449
1450 delete_vma_from_mm(vma);
1451 down_write(&nommu_region_sem);
1452 delete_nommu_region(vma->vm_region);
1453 if (new_below) {
1454 vma->vm_region->vm_start = vma->vm_start = addr;
1455 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1456 } else {
1457 vma->vm_region->vm_end = vma->vm_end = addr;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001458 vma->vm_region->vm_top = addr;
David Howells8feae132009-01-08 12:04:47 +00001459 }
1460 add_nommu_region(vma->vm_region);
1461 add_nommu_region(new->vm_region);
1462 up_write(&nommu_region_sem);
1463 add_vma_to_mm(mm, vma);
1464 add_vma_to_mm(mm, new);
1465 return 0;
1466}
1467
1468/*
1469 * shrink a VMA by removing the specified chunk from either the beginning or
1470 * the end
1471 */
1472static int shrink_vma(struct mm_struct *mm,
1473 struct vm_area_struct *vma,
1474 unsigned long from, unsigned long to)
1475{
1476 struct vm_region *region;
1477
1478 kenter("");
1479
1480 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1481 * and list */
1482 delete_vma_from_mm(vma);
1483 if (from > vma->vm_start)
1484 vma->vm_end = from;
1485 else
1486 vma->vm_start = to;
1487 add_vma_to_mm(mm, vma);
1488
1489 /* cut the backing region down to size */
1490 region = vma->vm_region;
1491 BUG_ON(atomic_read(&region->vm_usage) != 1);
1492
1493 down_write(&nommu_region_sem);
1494 delete_nommu_region(region);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001495 if (from > region->vm_start) {
1496 to = region->vm_top;
1497 region->vm_top = region->vm_end = from;
1498 } else {
David Howells8feae132009-01-08 12:04:47 +00001499 region->vm_start = to;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001500 }
David Howells8feae132009-01-08 12:04:47 +00001501 add_nommu_region(region);
1502 up_write(&nommu_region_sem);
1503
1504 free_page_series(from, to);
1505 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506}
1507
David Howells30340972006-09-27 01:50:20 -07001508/*
1509 * release a mapping
David Howells8feae132009-01-08 12:04:47 +00001510 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1511 * VMA, though it need not cover the whole VMA
David Howells30340972006-09-27 01:50:20 -07001512 */
David Howells8feae132009-01-08 12:04:47 +00001513int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514{
David Howells8feae132009-01-08 12:04:47 +00001515 struct vm_area_struct *vma;
1516 struct rb_node *rb;
1517 unsigned long end = start + len;
1518 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519
David Howells8feae132009-01-08 12:04:47 +00001520 kenter(",%lx,%zx", start, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521
David Howells8feae132009-01-08 12:04:47 +00001522 if (len == 0)
1523 return -EINVAL;
1524
1525 /* find the first potentially overlapping VMA */
1526 vma = find_vma(mm, start);
1527 if (!vma) {
David Howells33e5d7692009-04-02 16:56:32 -07001528 static int limit = 0;
1529 if (limit < 5) {
1530 printk(KERN_WARNING
1531 "munmap of memory not mmapped by process %d"
1532 " (%s): 0x%lx-0x%lx\n",
1533 current->pid, current->comm,
1534 start, start + len - 1);
1535 limit++;
1536 }
David Howells8feae132009-01-08 12:04:47 +00001537 return -EINVAL;
David Howells30340972006-09-27 01:50:20 -07001538 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539
David Howells8feae132009-01-08 12:04:47 +00001540 /* we're allowed to split an anonymous VMA but not a file-backed one */
1541 if (vma->vm_file) {
1542 do {
1543 if (start > vma->vm_start) {
1544 kleave(" = -EINVAL [miss]");
1545 return -EINVAL;
1546 }
1547 if (end == vma->vm_end)
1548 goto erase_whole_vma;
1549 rb = rb_next(&vma->vm_rb);
1550 vma = rb_entry(rb, struct vm_area_struct, vm_rb);
1551 } while (rb);
1552 kleave(" = -EINVAL [split file]");
1553 return -EINVAL;
1554 } else {
1555 /* the chunk must be a subset of the VMA found */
1556 if (start == vma->vm_start && end == vma->vm_end)
1557 goto erase_whole_vma;
1558 if (start < vma->vm_start || end > vma->vm_end) {
1559 kleave(" = -EINVAL [superset]");
1560 return -EINVAL;
1561 }
1562 if (start & ~PAGE_MASK) {
1563 kleave(" = -EINVAL [unaligned start]");
1564 return -EINVAL;
1565 }
1566 if (end != vma->vm_end && end & ~PAGE_MASK) {
1567 kleave(" = -EINVAL [unaligned split]");
1568 return -EINVAL;
1569 }
1570 if (start != vma->vm_start && end != vma->vm_end) {
1571 ret = split_vma(mm, vma, start, 1);
1572 if (ret < 0) {
1573 kleave(" = %d [split]", ret);
1574 return ret;
1575 }
1576 }
1577 return shrink_vma(mm, vma, start, end);
1578 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579
David Howells8feae132009-01-08 12:04:47 +00001580erase_whole_vma:
1581 delete_vma_from_mm(vma);
1582 delete_vma(mm, vma);
1583 kleave(" = 0");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584 return 0;
1585}
Paul Mundtb5073172007-07-21 04:37:25 -07001586EXPORT_SYMBOL(do_munmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587
Heiko Carstens6a6160a2009-01-14 14:14:15 +01001588SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
David Howells30340972006-09-27 01:50:20 -07001589{
1590 int ret;
1591 struct mm_struct *mm = current->mm;
1592
1593 down_write(&mm->mmap_sem);
1594 ret = do_munmap(mm, addr, len);
1595 up_write(&mm->mmap_sem);
1596 return ret;
1597}
1598
1599/*
David Howells8feae132009-01-08 12:04:47 +00001600 * release all the mappings made in a process's VM space
David Howells30340972006-09-27 01:50:20 -07001601 */
David Howells8feae132009-01-08 12:04:47 +00001602void exit_mmap(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603{
David Howells8feae132009-01-08 12:04:47 +00001604 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605
David Howells8feae132009-01-08 12:04:47 +00001606 if (!mm)
1607 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608
David Howells8feae132009-01-08 12:04:47 +00001609 kenter("");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610
David Howells8feae132009-01-08 12:04:47 +00001611 mm->total_vm = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612
David Howells8feae132009-01-08 12:04:47 +00001613 while ((vma = mm->mmap)) {
1614 mm->mmap = vma->vm_next;
1615 delete_vma_from_mm(vma);
1616 delete_vma(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617 }
David Howells8feae132009-01-08 12:04:47 +00001618
1619 kleave("");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620}
1621
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622unsigned long do_brk(unsigned long addr, unsigned long len)
1623{
1624 return -ENOMEM;
1625}
1626
1627/*
David Howells6fa5f802006-09-27 01:50:21 -07001628 * expand (or shrink) an existing mapping, potentially moving it at the same
1629 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630 *
David Howells6fa5f802006-09-27 01:50:21 -07001631 * under NOMMU conditions, we only permit changing a mapping's size, and only
David Howells8feae132009-01-08 12:04:47 +00001632 * as long as it stays within the region allocated by do_mmap_private() and the
1633 * block is not shareable
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634 *
David Howells6fa5f802006-09-27 01:50:21 -07001635 * MREMAP_FIXED is not supported under NOMMU conditions
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 */
1637unsigned long do_mremap(unsigned long addr,
1638 unsigned long old_len, unsigned long new_len,
1639 unsigned long flags, unsigned long new_addr)
1640{
David Howells6fa5f802006-09-27 01:50:21 -07001641 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642
1643 /* insanity checks first */
David Howells8feae132009-01-08 12:04:47 +00001644 if (old_len == 0 || new_len == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645 return (unsigned long) -EINVAL;
1646
David Howells8feae132009-01-08 12:04:47 +00001647 if (addr & ~PAGE_MASK)
1648 return -EINVAL;
1649
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650 if (flags & MREMAP_FIXED && new_addr != addr)
1651 return (unsigned long) -EINVAL;
1652
David Howells8feae132009-01-08 12:04:47 +00001653 vma = find_vma_exact(current->mm, addr, old_len);
David Howells6fa5f802006-09-27 01:50:21 -07001654 if (!vma)
1655 return (unsigned long) -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656
David Howells6fa5f802006-09-27 01:50:21 -07001657 if (vma->vm_end != vma->vm_start + old_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 return (unsigned long) -EFAULT;
1659
David Howells6fa5f802006-09-27 01:50:21 -07001660 if (vma->vm_flags & VM_MAYSHARE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 return (unsigned long) -EPERM;
1662
David Howells8feae132009-01-08 12:04:47 +00001663 if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 return (unsigned long) -ENOMEM;
1665
1666 /* all checks complete - do it */
David Howells6fa5f802006-09-27 01:50:21 -07001667 vma->vm_end = vma->vm_start + new_len;
David Howells6fa5f802006-09-27 01:50:21 -07001668 return vma->vm_start;
1669}
Paul Mundtb5073172007-07-21 04:37:25 -07001670EXPORT_SYMBOL(do_mremap);
David Howells6fa5f802006-09-27 01:50:21 -07001671
Heiko Carstens6a6160a2009-01-14 14:14:15 +01001672SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1673 unsigned long, new_len, unsigned long, flags,
1674 unsigned long, new_addr)
David Howells6fa5f802006-09-27 01:50:21 -07001675{
1676 unsigned long ret;
1677
1678 down_write(&current->mm->mmap_sem);
1679 ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1680 up_write(&current->mm->mmap_sem);
1681 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682}
1683
Linus Torvalds6aab3412005-11-28 14:34:23 -08001684struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001685 unsigned int foll_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686{
1687 return NULL;
1688}
1689
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690int remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
1691 unsigned long to, unsigned long size, pgprot_t prot)
1692{
Greg Ungerer66aa2b42005-09-12 11:18:10 +10001693 vma->vm_start = vma->vm_pgoff << PAGE_SHIFT;
1694 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695}
Luke Yang22c4af42006-07-14 00:24:09 -07001696EXPORT_SYMBOL(remap_pfn_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697
Paul Mundtf905bc42008-02-04 22:29:59 -08001698int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1699 unsigned long pgoff)
1700{
1701 unsigned int size = vma->vm_end - vma->vm_start;
1702
1703 if (!(vma->vm_flags & VM_USERMAP))
1704 return -EINVAL;
1705
1706 vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1707 vma->vm_end = vma->vm_start + size;
1708
1709 return 0;
1710}
1711EXPORT_SYMBOL(remap_vmalloc_range);
1712
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1714{
1715}
1716
1717unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1718 unsigned long len, unsigned long pgoff, unsigned long flags)
1719{
1720 return -ENOMEM;
1721}
1722
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001723void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724{
1725}
1726
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727void unmap_mapping_range(struct address_space *mapping,
1728 loff_t const holebegin, loff_t const holelen,
1729 int even_cows)
1730{
1731}
Luke Yang22c4af42006-07-14 00:24:09 -07001732EXPORT_SYMBOL(unmap_mapping_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733
1734/*
David Howellsd56e03c2007-03-22 00:11:23 -08001735 * ask for an unmapped area at which to create a mapping on a file
1736 */
1737unsigned long get_unmapped_area(struct file *file, unsigned long addr,
1738 unsigned long len, unsigned long pgoff,
1739 unsigned long flags)
1740{
1741 unsigned long (*get_area)(struct file *, unsigned long, unsigned long,
1742 unsigned long, unsigned long);
1743
1744 get_area = current->mm->get_unmapped_area;
1745 if (file && file->f_op && file->f_op->get_unmapped_area)
1746 get_area = file->f_op->get_unmapped_area;
1747
1748 if (!get_area)
1749 return -ENOSYS;
1750
1751 return get_area(file, addr, len, pgoff, flags);
1752}
David Howellsd56e03c2007-03-22 00:11:23 -08001753EXPORT_SYMBOL(get_unmapped_area);
1754
1755/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756 * Check that a process has enough memory to allocate a new virtual
1757 * mapping. 0 means there is enough memory for the allocation to
1758 * succeed and -ENOMEM implies there is not.
1759 *
1760 * We currently support three overcommit policies, which are set via the
1761 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
1762 *
1763 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1764 * Additional code 2002 Jul 20 by Robert Love.
1765 *
1766 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1767 *
1768 * Note this is a helper function intended to be used by LSMs which
1769 * wish to use this logic.
1770 */
Alan Cox34b4e4a2007-08-22 14:01:28 -07001771int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772{
1773 unsigned long free, allowed;
1774
1775 vm_acct_memory(pages);
1776
1777 /*
1778 * Sometimes we want to use more memory than we have
1779 */
1780 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1781 return 0;
1782
1783 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1784 unsigned long n;
1785
Christoph Lameter347ce432006-06-30 01:55:35 -07001786 free = global_page_state(NR_FILE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787 free += nr_swap_pages;
1788
1789 /*
1790 * Any slabs which are created with the
1791 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1792 * which are reclaimable, under pressure. The dentry
1793 * cache and most inode caches should fall into this
1794 */
Christoph Lameter972d1a72006-09-25 23:31:51 -07001795 free += global_page_state(NR_SLAB_RECLAIMABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796
1797 /*
1798 * Leave the last 3% for root
1799 */
1800 if (!cap_sys_admin)
1801 free -= free / 32;
1802
1803 if (free > pages)
1804 return 0;
1805
1806 /*
1807 * nr_free_pages() is very expensive on large systems,
1808 * only call if we're about to fail.
1809 */
1810 n = nr_free_pages();
Hideo AOKId5ddc792006-04-10 22:53:01 -07001811
1812 /*
1813 * Leave reserved pages. The pages are not for anonymous pages.
1814 */
1815 if (n <= totalreserve_pages)
1816 goto error;
1817 else
1818 n -= totalreserve_pages;
1819
1820 /*
1821 * Leave the last 3% for root
1822 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823 if (!cap_sys_admin)
1824 n -= n / 32;
1825 free += n;
1826
1827 if (free > pages)
1828 return 0;
Hideo AOKId5ddc792006-04-10 22:53:01 -07001829
1830 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831 }
1832
1833 allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1834 /*
1835 * Leave the last 3% for root
1836 */
1837 if (!cap_sys_admin)
1838 allowed -= allowed / 32;
1839 allowed += total_swap_pages;
1840
1841 /* Don't let a single process grow too big:
1842 leave 3% of the size of this process for other processes */
Alan Cox731572d2008-10-29 14:01:20 -07001843 if (mm)
1844 allowed -= mm->total_vm / 32;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07001846 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847 return 0;
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07001848
Hideo AOKId5ddc792006-04-10 22:53:01 -07001849error:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850 vm_unacct_memory(pages);
1851
1852 return -ENOMEM;
1853}
1854
1855int in_gate_area_no_task(unsigned long addr)
1856{
1857 return 0;
1858}
David Howellsb0e15192006-01-06 00:11:42 -08001859
Nick Piggind0217ac2007-07-19 01:47:03 -07001860int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
David Howellsb0e15192006-01-06 00:11:42 -08001861{
1862 BUG();
Nick Piggind0217ac2007-07-19 01:47:03 -07001863 return 0;
David Howellsb0e15192006-01-06 00:11:42 -08001864}
Paul Mundtb5073172007-07-21 04:37:25 -07001865EXPORT_SYMBOL(filemap_fault);
David Howells0ec76a12006-09-27 01:50:15 -07001866
1867/*
1868 * Access another process' address space.
1869 * - source/target buffer must be kernel space
1870 */
1871int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
1872{
David Howells0ec76a12006-09-27 01:50:15 -07001873 struct vm_area_struct *vma;
1874 struct mm_struct *mm;
1875
1876 if (addr + len < addr)
1877 return 0;
1878
1879 mm = get_task_mm(tsk);
1880 if (!mm)
1881 return 0;
1882
1883 down_read(&mm->mmap_sem);
1884
1885 /* the access must start within one of the target process's mappings */
David Howells0159b142006-09-27 01:50:16 -07001886 vma = find_vma(mm, addr);
1887 if (vma) {
David Howells0ec76a12006-09-27 01:50:15 -07001888 /* don't overrun this mapping */
1889 if (addr + len >= vma->vm_end)
1890 len = vma->vm_end - addr;
1891
1892 /* only read or write mappings where it is permitted */
David Howellsd00c7b992006-09-27 01:50:19 -07001893 if (write && vma->vm_flags & VM_MAYWRITE)
David Howells0ec76a12006-09-27 01:50:15 -07001894 len -= copy_to_user((void *) addr, buf, len);
David Howellsd00c7b992006-09-27 01:50:19 -07001895 else if (!write && vma->vm_flags & VM_MAYREAD)
David Howells0ec76a12006-09-27 01:50:15 -07001896 len -= copy_from_user(buf, (void *) addr, len);
1897 else
1898 len = 0;
1899 } else {
1900 len = 0;
1901 }
1902
1903 up_read(&mm->mmap_sem);
1904 mmput(mm);
1905 return len;
1906}