blob: 809998aa7b509dd072862def05f420a14e1185cf [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>
David Howells8feae132009-01-08 12:04:47 +000036#include "internal.h"
37
38static inline __attribute__((format(printf, 1, 2)))
39void no_printk(const char *fmt, ...)
40{
41}
42
43#if 0
44#define kenter(FMT, ...) \
45 printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
46#define kleave(FMT, ...) \
47 printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
48#define kdebug(FMT, ...) \
49 printk(KERN_DEBUG "xxx" FMT"yyy\n", ##__VA_ARGS__)
50#else
51#define kenter(FMT, ...) \
52 no_printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
53#define kleave(FMT, ...) \
54 no_printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
55#define kdebug(FMT, ...) \
56 no_printk(KERN_DEBUG FMT"\n", ##__VA_ARGS__)
57#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -070058
Nick Pigginb291f002008-10-18 20:26:44 -070059#include "internal.h"
60
Linus Torvalds1da177e2005-04-16 15:20:36 -070061void *high_memory;
62struct page *mem_map;
63unsigned long max_mapnr;
64unsigned long num_physpages;
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;
Paul Mundtdd8632a2009-01-08 12:04:47 +000069int sysctl_nr_trim_pages = 1; /* page trimming behaviour */
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
82struct vm_operations_struct generic_file_vm_ops = {
83};
84
85/*
86 * Handle all mappings that got truncated by a "truncate()"
87 * system call.
88 *
89 * NOTE! We have to be ready to update the memory sharing
90 * between the file and the memory map for a potential last
91 * incomplete page. Ugly, but necessary.
92 */
93int vmtruncate(struct inode *inode, loff_t offset)
94{
95 struct address_space *mapping = inode->i_mapping;
96 unsigned long limit;
97
98 if (inode->i_size < offset)
99 goto do_expand;
100 i_size_write(inode, offset);
101
102 truncate_inode_pages(mapping, offset);
103 goto out_truncate;
104
105do_expand:
106 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
107 if (limit != RLIM_INFINITY && offset > limit)
108 goto out_sig;
109 if (offset > inode->i_sb->s_maxbytes)
110 goto out;
111 i_size_write(inode, offset);
112
113out_truncate:
Al Viroacfa4382008-12-04 10:06:33 -0500114 if (inode->i_op->truncate)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115 inode->i_op->truncate(inode);
116 return 0;
117out_sig:
118 send_sig(SIGXFSZ, current, 0);
119out:
120 return -EFBIG;
121}
122
123EXPORT_SYMBOL(vmtruncate);
124
125/*
126 * Return the total memory allocated for this pointer, not
127 * just what the caller asked for.
128 *
129 * Doesn't have to be accurate, i.e. may have races.
130 */
131unsigned int kobjsize(const void *objp)
132{
133 struct page *page;
134
Michael Hennerich4016a132008-04-28 02:13:38 -0700135 /*
136 * If the object we have should not have ksize performed on it,
137 * return size of 0
138 */
Paul Mundt5a1603b2008-06-12 16:29:55 +0900139 if (!objp || !virt_addr_valid(objp))
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700140 return 0;
141
142 page = virt_to_head_page(objp);
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700143
144 /*
145 * If the allocator sets PageSlab, we know the pointer came from
146 * kmalloc().
147 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148 if (PageSlab(page))
149 return ksize(objp);
150
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700151 /*
Paul Mundtab2e83e2009-01-08 12:04:48 +0000152 * If it's not a compound page, see if we have a matching VMA
153 * region. This test is intentionally done in reverse order,
154 * so if there's no VMA, we still fall through and hand back
155 * PAGE_SIZE for 0-order pages.
156 */
157 if (!PageCompound(page)) {
158 struct vm_area_struct *vma;
159
160 vma = find_vma(current->mm, (unsigned long)objp);
161 if (vma)
162 return vma->vm_end - vma->vm_start;
163 }
164
165 /*
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700166 * The ksize() function is only guaranteed to work for pointers
Paul Mundt5a1603b2008-06-12 16:29:55 +0900167 * returned by kmalloc(). So handle arbitrary pointers here.
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700168 */
Paul Mundt5a1603b2008-06-12 16:29:55 +0900169 return PAGE_SIZE << compound_order(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170}
171
Nick Pigginb291f002008-10-18 20:26:44 -0700172int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
173 unsigned long start, int len, int flags,
174 struct page **pages, struct vm_area_struct **vmas)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175{
Sonic Zhang910e46d2006-09-27 01:50:17 -0700176 struct vm_area_struct *vma;
David Howells7b4d5b82006-09-27 01:50:18 -0700177 unsigned long vm_flags;
178 int i;
Nick Pigginb291f002008-10-18 20:26:44 -0700179 int write = !!(flags & GUP_FLAGS_WRITE);
180 int force = !!(flags & GUP_FLAGS_FORCE);
181 int ignore = !!(flags & GUP_FLAGS_IGNORE_VMA_PERMISSIONS);
David Howells7b4d5b82006-09-27 01:50:18 -0700182
183 /* calculate required read or write permissions.
184 * - if 'force' is set, we only require the "MAY" flags.
185 */
186 vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
187 vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188
189 for (i = 0; i < len; i++) {
Sonic Zhang910e46d2006-09-27 01:50:17 -0700190 vma = find_vma(mm, start);
David Howells7b4d5b82006-09-27 01:50:18 -0700191 if (!vma)
192 goto finish_or_fault;
193
194 /* protect what we can, including chardevs */
195 if (vma->vm_flags & (VM_IO | VM_PFNMAP) ||
Nick Pigginb291f002008-10-18 20:26:44 -0700196 (!ignore && !(vm_flags & vma->vm_flags)))
David Howells7b4d5b82006-09-27 01:50:18 -0700197 goto finish_or_fault;
Sonic Zhang910e46d2006-09-27 01:50:17 -0700198
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 if (pages) {
200 pages[i] = virt_to_page(start);
201 if (pages[i])
202 page_cache_get(pages[i]);
203 }
204 if (vmas)
Sonic Zhang910e46d2006-09-27 01:50:17 -0700205 vmas[i] = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 start += PAGE_SIZE;
207 }
David Howells7b4d5b82006-09-27 01:50:18 -0700208
209 return i;
210
211finish_or_fault:
212 return i ? : -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213}
Nick Pigginb291f002008-10-18 20:26:44 -0700214
215
216/*
217 * get a list of pages in an address range belonging to the specified process
218 * and indicate the VMA that covers each page
219 * - this is potentially dodgy as we may end incrementing the page count of a
220 * slab page or a secondary page from a compound page
221 * - don't permit access to VMAs that don't support it, such as I/O mappings
222 */
223int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
224 unsigned long start, int len, int write, int force,
225 struct page **pages, struct vm_area_struct **vmas)
226{
227 int flags = 0;
228
229 if (write)
230 flags |= GUP_FLAGS_WRITE;
231 if (force)
232 flags |= GUP_FLAGS_FORCE;
233
234 return __get_user_pages(tsk, mm,
235 start, len, flags,
236 pages, vmas);
237}
Greg Ungerer66aa2b42005-09-12 11:18:10 +1000238EXPORT_SYMBOL(get_user_pages);
239
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240DEFINE_RWLOCK(vmlist_lock);
241struct vm_struct *vmlist;
242
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800243void vfree(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244{
245 kfree(addr);
246}
Paul Mundtb5073172007-07-21 04:37:25 -0700247EXPORT_SYMBOL(vfree);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248
Al Virodd0fc662005-10-07 07:46:04 +0100249void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250{
251 /*
Robert P. J. Day85186092007-10-19 23:11:38 +0200252 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
253 * returns only a logical address.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 */
Nick Piggin84097512006-03-22 00:08:34 -0800255 return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256}
Paul Mundtb5073172007-07-21 04:37:25 -0700257EXPORT_SYMBOL(__vmalloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258
Paul Mundtf905bc42008-02-04 22:29:59 -0800259void *vmalloc_user(unsigned long size)
260{
261 void *ret;
262
263 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
264 PAGE_KERNEL);
265 if (ret) {
266 struct vm_area_struct *vma;
267
268 down_write(&current->mm->mmap_sem);
269 vma = find_vma(current->mm, (unsigned long)ret);
270 if (vma)
271 vma->vm_flags |= VM_USERMAP;
272 up_write(&current->mm->mmap_sem);
273 }
274
275 return ret;
276}
277EXPORT_SYMBOL(vmalloc_user);
278
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800279struct page *vmalloc_to_page(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280{
281 return virt_to_page(addr);
282}
Paul Mundtb5073172007-07-21 04:37:25 -0700283EXPORT_SYMBOL(vmalloc_to_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800285unsigned long vmalloc_to_pfn(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286{
287 return page_to_pfn(virt_to_page(addr));
288}
Paul Mundtb5073172007-07-21 04:37:25 -0700289EXPORT_SYMBOL(vmalloc_to_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290
291long vread(char *buf, char *addr, unsigned long count)
292{
293 memcpy(buf, addr, count);
294 return count;
295}
296
297long vwrite(char *buf, char *addr, unsigned long count)
298{
299 /* Don't allow overflow */
300 if ((unsigned long) addr + count < count)
301 count = -(unsigned long) addr;
302
303 memcpy(addr, buf, count);
304 return(count);
305}
306
307/*
308 * vmalloc - allocate virtually continguos memory
309 *
310 * @size: allocation size
311 *
312 * Allocate enough pages to cover @size from the page level
313 * allocator and map them into continguos kernel virtual space.
314 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +0200315 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 * use __vmalloc() instead.
317 */
318void *vmalloc(unsigned long size)
319{
320 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
321}
Andrew Mortonf6138882006-02-28 16:59:18 -0800322EXPORT_SYMBOL(vmalloc);
323
324void *vmalloc_node(unsigned long size, int node)
325{
326 return vmalloc(size);
327}
328EXPORT_SYMBOL(vmalloc_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329
Paul Mundt1af446e2008-08-04 16:01:47 +0900330#ifndef PAGE_KERNEL_EXEC
331# define PAGE_KERNEL_EXEC PAGE_KERNEL
332#endif
333
334/**
335 * vmalloc_exec - allocate virtually contiguous, executable memory
336 * @size: allocation size
337 *
338 * Kernel-internal function to allocate enough pages to cover @size
339 * the page level allocator and map them into contiguous and
340 * executable kernel virtual space.
341 *
342 * For tight control over page level allocator and protection flags
343 * use __vmalloc() instead.
344 */
345
346void *vmalloc_exec(unsigned long size)
347{
348 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
349}
350
Paul Mundtb5073172007-07-21 04:37:25 -0700351/**
352 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 * @size: allocation size
354 *
355 * Allocate enough 32bit PA addressable pages to cover @size from the
356 * page level allocator and map them into continguos kernel virtual space.
357 */
358void *vmalloc_32(unsigned long size)
359{
360 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
361}
Paul Mundtb5073172007-07-21 04:37:25 -0700362EXPORT_SYMBOL(vmalloc_32);
363
364/**
365 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
366 * @size: allocation size
367 *
368 * The resulting memory area is 32bit addressable and zeroed so it can be
369 * mapped to userspace without leaking data.
Paul Mundtf905bc42008-02-04 22:29:59 -0800370 *
371 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
372 * remap_vmalloc_range() are permissible.
Paul Mundtb5073172007-07-21 04:37:25 -0700373 */
374void *vmalloc_32_user(unsigned long size)
375{
Paul Mundtf905bc42008-02-04 22:29:59 -0800376 /*
377 * We'll have to sort out the ZONE_DMA bits for 64-bit,
378 * but for now this can simply use vmalloc_user() directly.
379 */
380 return vmalloc_user(size);
Paul Mundtb5073172007-07-21 04:37:25 -0700381}
382EXPORT_SYMBOL(vmalloc_32_user);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383
384void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
385{
386 BUG();
387 return NULL;
388}
Paul Mundtb5073172007-07-21 04:37:25 -0700389EXPORT_SYMBOL(vmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800391void vunmap(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392{
393 BUG();
394}
Paul Mundtb5073172007-07-21 04:37:25 -0700395EXPORT_SYMBOL(vunmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396
Paul Mundteb6434d2009-01-21 17:45:47 +0900397void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
398{
399 BUG();
400 return NULL;
401}
402EXPORT_SYMBOL(vm_map_ram);
403
404void vm_unmap_ram(const void *mem, unsigned int count)
405{
406 BUG();
407}
408EXPORT_SYMBOL(vm_unmap_ram);
409
410void vm_unmap_aliases(void)
411{
412}
413EXPORT_SYMBOL_GPL(vm_unmap_aliases);
414
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415/*
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -0700416 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
417 * have one.
418 */
419void __attribute__((weak)) vmalloc_sync_all(void)
420{
421}
422
Paul Mundtb5073172007-07-21 04:37:25 -0700423int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
424 struct page *page)
425{
426 return -EINVAL;
427}
428EXPORT_SYMBOL(vm_insert_page);
429
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -0700430/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 * sys_brk() for the most part doesn't need the global kernel
432 * lock, except when an application is doing something nasty
433 * like trying to un-brk an area that has already been mapped
434 * to a regular file. in this case, the unmapping will need
435 * to invoke file system routines that need the global lock.
436 */
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100437SYSCALL_DEFINE1(brk, unsigned long, brk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438{
439 struct mm_struct *mm = current->mm;
440
441 if (brk < mm->start_brk || brk > mm->context.end_brk)
442 return mm->brk;
443
444 if (mm->brk == brk)
445 return mm->brk;
446
447 /*
448 * Always allow shrinking brk
449 */
450 if (brk <= mm->brk) {
451 mm->brk = brk;
452 return brk;
453 }
454
455 /*
456 * Ok, looks good - let it rip.
457 */
458 return mm->brk = brk;
459}
460
David Howells8feae132009-01-08 12:04:47 +0000461/*
462 * initialise the VMA and region record slabs
463 */
464void __init mmap_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465{
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -0700466 int ret;
467
468 ret = percpu_counter_init(&vm_committed_as, 0);
469 VM_BUG_ON(ret);
David Howells33e5d7692009-04-02 16:56:32 -0700470 vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC);
David Howells8feae132009-01-08 12:04:47 +0000471}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472
David Howells8feae132009-01-08 12:04:47 +0000473/*
474 * validate the region tree
475 * - the caller must hold the region lock
476 */
477#ifdef CONFIG_DEBUG_NOMMU_REGIONS
478static noinline void validate_nommu_regions(void)
479{
480 struct vm_region *region, *last;
481 struct rb_node *p, *lastp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482
David Howells8feae132009-01-08 12:04:47 +0000483 lastp = rb_first(&nommu_region_tree);
484 if (!lastp)
485 return;
486
487 last = rb_entry(lastp, struct vm_region, vm_rb);
David Howells33e5d7692009-04-02 16:56:32 -0700488 BUG_ON(unlikely(last->vm_end <= last->vm_start));
489 BUG_ON(unlikely(last->vm_top < last->vm_end));
David Howells8feae132009-01-08 12:04:47 +0000490
491 while ((p = rb_next(lastp))) {
492 region = rb_entry(p, struct vm_region, vm_rb);
493 last = rb_entry(lastp, struct vm_region, vm_rb);
494
David Howells33e5d7692009-04-02 16:56:32 -0700495 BUG_ON(unlikely(region->vm_end <= region->vm_start));
496 BUG_ON(unlikely(region->vm_top < region->vm_end));
497 BUG_ON(unlikely(region->vm_start < last->vm_top));
David Howells8feae132009-01-08 12:04:47 +0000498
499 lastp = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 }
501}
David Howells8feae132009-01-08 12:04:47 +0000502#else
David Howells33e5d7692009-04-02 16:56:32 -0700503static void validate_nommu_regions(void)
504{
505}
David Howells8feae132009-01-08 12:04:47 +0000506#endif
507
508/*
509 * add a region into the global tree
510 */
511static void add_nommu_region(struct vm_region *region)
512{
513 struct vm_region *pregion;
514 struct rb_node **p, *parent;
515
516 validate_nommu_regions();
517
518 BUG_ON(region->vm_start & ~PAGE_MASK);
519
520 parent = NULL;
521 p = &nommu_region_tree.rb_node;
522 while (*p) {
523 parent = *p;
524 pregion = rb_entry(parent, struct vm_region, vm_rb);
525 if (region->vm_start < pregion->vm_start)
526 p = &(*p)->rb_left;
527 else if (region->vm_start > pregion->vm_start)
528 p = &(*p)->rb_right;
529 else if (pregion == region)
530 return;
531 else
532 BUG();
533 }
534
535 rb_link_node(&region->vm_rb, parent, p);
536 rb_insert_color(&region->vm_rb, &nommu_region_tree);
537
538 validate_nommu_regions();
539}
540
541/*
542 * delete a region from the global tree
543 */
544static void delete_nommu_region(struct vm_region *region)
545{
546 BUG_ON(!nommu_region_tree.rb_node);
547
548 validate_nommu_regions();
549 rb_erase(&region->vm_rb, &nommu_region_tree);
550 validate_nommu_regions();
551}
552
553/*
554 * free a contiguous series of pages
555 */
556static void free_page_series(unsigned long from, unsigned long to)
557{
558 for (; from < to; from += PAGE_SIZE) {
559 struct page *page = virt_to_page(from);
560
561 kdebug("- free %lx", from);
David Howells33e5d7692009-04-02 16:56:32 -0700562 atomic_long_dec(&mmap_pages_allocated);
David Howells8feae132009-01-08 12:04:47 +0000563 if (page_count(page) != 1)
David Howells33e5d7692009-04-02 16:56:32 -0700564 kdebug("free page %p: refcount not one: %d",
565 page, page_count(page));
David Howells8feae132009-01-08 12:04:47 +0000566 put_page(page);
567 }
568}
569
570/*
571 * release a reference to a region
David Howells33e5d7692009-04-02 16:56:32 -0700572 * - the caller must hold the region semaphore for writing, which this releases
Paul Mundtdd8632a2009-01-08 12:04:47 +0000573 * - the region may not have been added to the tree yet, in which case vm_top
David Howells8feae132009-01-08 12:04:47 +0000574 * will equal vm_start
575 */
576static void __put_nommu_region(struct vm_region *region)
577 __releases(nommu_region_sem)
578{
579 kenter("%p{%d}", region, atomic_read(&region->vm_usage));
580
581 BUG_ON(!nommu_region_tree.rb_node);
582
583 if (atomic_dec_and_test(&region->vm_usage)) {
Paul Mundtdd8632a2009-01-08 12:04:47 +0000584 if (region->vm_top > region->vm_start)
David Howells8feae132009-01-08 12:04:47 +0000585 delete_nommu_region(region);
586 up_write(&nommu_region_sem);
587
588 if (region->vm_file)
589 fput(region->vm_file);
590
591 /* IO memory and memory shared directly out of the pagecache
592 * from ramfs/tmpfs mustn't be released here */
593 if (region->vm_flags & VM_MAPPED_COPY) {
594 kdebug("free series");
Paul Mundtdd8632a2009-01-08 12:04:47 +0000595 free_page_series(region->vm_start, region->vm_top);
David Howells8feae132009-01-08 12:04:47 +0000596 }
597 kmem_cache_free(vm_region_jar, region);
598 } else {
599 up_write(&nommu_region_sem);
600 }
601}
602
603/*
604 * release a reference to a region
605 */
606static void put_nommu_region(struct vm_region *region)
607{
608 down_write(&nommu_region_sem);
609 __put_nommu_region(region);
610}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611
David Howells30340972006-09-27 01:50:20 -0700612/*
613 * add a VMA into a process's mm_struct in the appropriate place in the list
David Howells8feae132009-01-08 12:04:47 +0000614 * and tree and add to the address space's page tree also if not an anonymous
615 * page
David Howells30340972006-09-27 01:50:20 -0700616 * - should be called with mm->mmap_sem held writelocked
617 */
David Howells8feae132009-01-08 12:04:47 +0000618static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
David Howells30340972006-09-27 01:50:20 -0700619{
David Howells8feae132009-01-08 12:04:47 +0000620 struct vm_area_struct *pvma, **pp;
621 struct address_space *mapping;
622 struct rb_node **p, *parent;
David Howells30340972006-09-27 01:50:20 -0700623
David Howells8feae132009-01-08 12:04:47 +0000624 kenter(",%p", vma);
625
626 BUG_ON(!vma->vm_region);
627
628 mm->map_count++;
629 vma->vm_mm = mm;
630
631 /* add the VMA to the mapping */
632 if (vma->vm_file) {
633 mapping = vma->vm_file->f_mapping;
634
635 flush_dcache_mmap_lock(mapping);
636 vma_prio_tree_insert(vma, &mapping->i_mmap);
637 flush_dcache_mmap_unlock(mapping);
638 }
639
640 /* add the VMA to the tree */
641 parent = NULL;
642 p = &mm->mm_rb.rb_node;
643 while (*p) {
644 parent = *p;
645 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
646
647 /* sort by: start addr, end addr, VMA struct addr in that order
648 * (the latter is necessary as we may get identical VMAs) */
649 if (vma->vm_start < pvma->vm_start)
650 p = &(*p)->rb_left;
651 else if (vma->vm_start > pvma->vm_start)
652 p = &(*p)->rb_right;
653 else if (vma->vm_end < pvma->vm_end)
654 p = &(*p)->rb_left;
655 else if (vma->vm_end > pvma->vm_end)
656 p = &(*p)->rb_right;
657 else if (vma < pvma)
658 p = &(*p)->rb_left;
659 else if (vma > pvma)
660 p = &(*p)->rb_right;
661 else
662 BUG();
663 }
664
665 rb_link_node(&vma->vm_rb, parent, p);
666 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
667
668 /* add VMA to the VMA list also */
669 for (pp = &mm->mmap; (pvma = *pp); pp = &(*pp)->vm_next) {
670 if (pvma->vm_start > vma->vm_start)
David Howells30340972006-09-27 01:50:20 -0700671 break;
David Howells8feae132009-01-08 12:04:47 +0000672 if (pvma->vm_start < vma->vm_start)
673 continue;
674 if (pvma->vm_end < vma->vm_end)
675 break;
676 }
David Howells30340972006-09-27 01:50:20 -0700677
David Howells8feae132009-01-08 12:04:47 +0000678 vma->vm_next = *pp;
679 *pp = vma;
680}
681
682/*
683 * delete a VMA from its owning mm_struct and address space
684 */
685static void delete_vma_from_mm(struct vm_area_struct *vma)
686{
687 struct vm_area_struct **pp;
688 struct address_space *mapping;
689 struct mm_struct *mm = vma->vm_mm;
690
691 kenter("%p", vma);
692
693 mm->map_count--;
694 if (mm->mmap_cache == vma)
695 mm->mmap_cache = NULL;
696
697 /* remove the VMA from the mapping */
698 if (vma->vm_file) {
699 mapping = vma->vm_file->f_mapping;
700
701 flush_dcache_mmap_lock(mapping);
702 vma_prio_tree_remove(vma, &mapping->i_mmap);
703 flush_dcache_mmap_unlock(mapping);
704 }
705
706 /* remove from the MM's tree and list */
707 rb_erase(&vma->vm_rb, &mm->mm_rb);
708 for (pp = &mm->mmap; *pp; pp = &(*pp)->vm_next) {
709 if (*pp == vma) {
710 *pp = vma->vm_next;
711 break;
712 }
713 }
714
715 vma->vm_mm = NULL;
716}
717
718/*
719 * destroy a VMA record
720 */
721static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
722{
723 kenter("%p", vma);
724 if (vma->vm_ops && vma->vm_ops->close)
725 vma->vm_ops->close(vma);
726 if (vma->vm_file) {
727 fput(vma->vm_file);
728 if (vma->vm_flags & VM_EXECUTABLE)
729 removed_exe_file_vma(mm);
730 }
731 put_nommu_region(vma->vm_region);
732 kmem_cache_free(vm_area_cachep, vma);
David Howells30340972006-09-27 01:50:20 -0700733}
734
735/*
736 * look up the first VMA in which addr resides, NULL if none
737 * - should be called with mm->mmap_sem at least held readlocked
738 */
739struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
740{
David Howells8feae132009-01-08 12:04:47 +0000741 struct vm_area_struct *vma;
742 struct rb_node *n = mm->mm_rb.rb_node;
David Howells30340972006-09-27 01:50:20 -0700743
David Howells8feae132009-01-08 12:04:47 +0000744 /* check the cache first */
745 vma = mm->mmap_cache;
746 if (vma && vma->vm_start <= addr && vma->vm_end > addr)
747 return vma;
748
749 /* trawl the tree (there may be multiple mappings in which addr
750 * resides) */
751 for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) {
752 vma = rb_entry(n, struct vm_area_struct, vm_rb);
753 if (vma->vm_start > addr)
754 return NULL;
755 if (vma->vm_end > addr) {
756 mm->mmap_cache = vma;
757 return vma;
758 }
David Howells30340972006-09-27 01:50:20 -0700759 }
760
David Howells30340972006-09-27 01:50:20 -0700761 return NULL;
762}
763EXPORT_SYMBOL(find_vma);
764
765/*
David Howells930e6522006-09-27 01:50:22 -0700766 * find a VMA
767 * - we don't extend stack VMAs under NOMMU conditions
768 */
769struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
770{
771 return find_vma(mm, addr);
772}
773
David Howells8feae132009-01-08 12:04:47 +0000774/*
775 * expand a stack to a given address
776 * - not supported under NOMMU conditions
777 */
Greg Ungerer57c8f632007-07-15 23:38:28 -0700778int expand_stack(struct vm_area_struct *vma, unsigned long address)
779{
780 return -ENOMEM;
781}
782
David Howells930e6522006-09-27 01:50:22 -0700783/*
David Howells6fa5f802006-09-27 01:50:21 -0700784 * look up the first VMA exactly that exactly matches addr
785 * - should be called with mm->mmap_sem at least held readlocked
786 */
David Howells8feae132009-01-08 12:04:47 +0000787static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
788 unsigned long addr,
789 unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790{
791 struct vm_area_struct *vma;
David Howells8feae132009-01-08 12:04:47 +0000792 struct rb_node *n = mm->mm_rb.rb_node;
793 unsigned long end = addr + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794
David Howells8feae132009-01-08 12:04:47 +0000795 /* check the cache first */
796 vma = mm->mmap_cache;
797 if (vma && vma->vm_start == addr && vma->vm_end == end)
798 return vma;
799
800 /* trawl the tree (there may be multiple mappings in which addr
801 * resides) */
802 for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 vma = rb_entry(n, struct vm_area_struct, vm_rb);
David Howells8feae132009-01-08 12:04:47 +0000804 if (vma->vm_start < addr)
805 continue;
806 if (vma->vm_start > addr)
807 return NULL;
808 if (vma->vm_end == end) {
809 mm->mmap_cache = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 return vma;
David Howells8feae132009-01-08 12:04:47 +0000811 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 }
813
814 return NULL;
815}
816
David Howells30340972006-09-27 01:50:20 -0700817/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 * determine whether a mapping should be permitted and, if so, what sort of
819 * mapping we're capable of supporting
820 */
821static int validate_mmap_request(struct file *file,
822 unsigned long addr,
823 unsigned long len,
824 unsigned long prot,
825 unsigned long flags,
826 unsigned long pgoff,
827 unsigned long *_capabilities)
828{
David Howells8feae132009-01-08 12:04:47 +0000829 unsigned long capabilities, rlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 unsigned long reqprot = prot;
831 int ret;
832
833 /* do the simple checks first */
834 if (flags & MAP_FIXED || addr) {
835 printk(KERN_DEBUG
836 "%d: Can't do fixed-address/overlay mmap of RAM\n",
837 current->pid);
838 return -EINVAL;
839 }
840
841 if ((flags & MAP_TYPE) != MAP_PRIVATE &&
842 (flags & MAP_TYPE) != MAP_SHARED)
843 return -EINVAL;
844
Mike Frysingerf81cff02006-12-06 12:02:59 +1000845 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846 return -EINVAL;
847
Mike Frysingerf81cff02006-12-06 12:02:59 +1000848 /* Careful about overflows.. */
David Howells8feae132009-01-08 12:04:47 +0000849 rlen = PAGE_ALIGN(len);
850 if (!rlen || rlen > TASK_SIZE)
Mike Frysingerf81cff02006-12-06 12:02:59 +1000851 return -ENOMEM;
852
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 /* offset overflow? */
David Howells8feae132009-01-08 12:04:47 +0000854 if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
Mike Frysingerf81cff02006-12-06 12:02:59 +1000855 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856
857 if (file) {
858 /* validate file mapping requests */
859 struct address_space *mapping;
860
861 /* files must support mmap */
862 if (!file->f_op || !file->f_op->mmap)
863 return -ENODEV;
864
865 /* work out if what we've got could possibly be shared
866 * - we support chardevs that provide their own "memory"
867 * - we support files/blockdevs that are memory backed
868 */
869 mapping = file->f_mapping;
870 if (!mapping)
Josef Sipeke9536ae2006-12-08 02:37:21 -0800871 mapping = file->f_path.dentry->d_inode->i_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872
873 capabilities = 0;
874 if (mapping && mapping->backing_dev_info)
875 capabilities = mapping->backing_dev_info->capabilities;
876
877 if (!capabilities) {
878 /* no explicit capabilities set, so assume some
879 * defaults */
Josef Sipeke9536ae2006-12-08 02:37:21 -0800880 switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 case S_IFREG:
882 case S_IFBLK:
883 capabilities = BDI_CAP_MAP_COPY;
884 break;
885
886 case S_IFCHR:
887 capabilities =
888 BDI_CAP_MAP_DIRECT |
889 BDI_CAP_READ_MAP |
890 BDI_CAP_WRITE_MAP;
891 break;
892
893 default:
894 return -EINVAL;
895 }
896 }
897
898 /* eliminate any capabilities that we can't support on this
899 * device */
900 if (!file->f_op->get_unmapped_area)
901 capabilities &= ~BDI_CAP_MAP_DIRECT;
902 if (!file->f_op->read)
903 capabilities &= ~BDI_CAP_MAP_COPY;
904
905 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;
1038 return ret;
1039 }
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,
1055 unsigned long len)
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 */
1066 if (vma->vm_file) {
1067 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;
1072 return ret;
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
1146 } else {
1147 /* if it's an anonymous mapping, then just clear it */
David Howells8feae132009-01-08 12:04:47 +00001148 memset(base, 0, rlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 }
1150
1151 return 0;
1152
1153error_free:
David Howells8feae132009-01-08 12:04:47 +00001154 free_page_series(region->vm_start, region->vm_end);
1155 region->vm_start = vma->vm_start = 0;
1156 region->vm_end = vma->vm_end = 0;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001157 region->vm_top = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 return ret;
1159
1160enomem:
Greg Ungerer05ae6fa2009-01-13 17:30:22 +10001161 printk("Allocation of length %lu from process %d (%s) failed\n",
1162 len, current->pid, current->comm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163 show_free_areas();
1164 return -ENOMEM;
1165}
1166
1167/*
1168 * handle mapping creation for uClinux
1169 */
1170unsigned long do_mmap_pgoff(struct file *file,
1171 unsigned long addr,
1172 unsigned long len,
1173 unsigned long prot,
1174 unsigned long flags,
1175 unsigned long pgoff)
1176{
David Howells8feae132009-01-08 12:04:47 +00001177 struct vm_area_struct *vma;
1178 struct vm_region *region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 struct rb_node *rb;
David Howells8feae132009-01-08 12:04:47 +00001180 unsigned long capabilities, vm_flags, result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181 int ret;
1182
David Howells8feae132009-01-08 12:04:47 +00001183 kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff);
1184
Eric Paris7cd94142007-11-26 18:47:40 -05001185 if (!(flags & MAP_FIXED))
1186 addr = round_hint_to_min(addr);
1187
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 /* decide whether we should attempt the mapping, and if so what sort of
1189 * mapping */
1190 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1191 &capabilities);
David Howells8feae132009-01-08 12:04:47 +00001192 if (ret < 0) {
1193 kleave(" = %d [val]", ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 return ret;
David Howells8feae132009-01-08 12:04:47 +00001195 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196
1197 /* we've determined that we can make the mapping, now translate what we
1198 * now know into VMA flags */
1199 vm_flags = determine_vm_flags(file, prot, flags, capabilities);
1200
David Howells8feae132009-01-08 12:04:47 +00001201 /* we're going to need to record the mapping */
1202 region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1203 if (!region)
1204 goto error_getting_region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205
David Howells8feae132009-01-08 12:04:47 +00001206 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1207 if (!vma)
1208 goto error_getting_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209
David Howells8feae132009-01-08 12:04:47 +00001210 atomic_set(&region->vm_usage, 1);
1211 region->vm_flags = vm_flags;
1212 region->vm_pgoff = pgoff;
1213
1214 INIT_LIST_HEAD(&vma->anon_vma_node);
1215 vma->vm_flags = vm_flags;
1216 vma->vm_pgoff = pgoff;
1217
1218 if (file) {
1219 region->vm_file = file;
1220 get_file(file);
1221 vma->vm_file = file;
1222 get_file(file);
1223 if (vm_flags & VM_EXECUTABLE) {
1224 added_exe_file_vma(current->mm);
1225 vma->vm_mm = current->mm;
1226 }
1227 }
1228
1229 down_write(&nommu_region_sem);
1230
1231 /* if we want to share, we need to check for regions created by other
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 * mmap() calls that overlap with our proposed mapping
David Howells8feae132009-01-08 12:04:47 +00001233 * - we can only share with a superset match on most regular files
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 * - shared mappings on character devices and memory backed files are
1235 * permitted to overlap inexactly as far as we are concerned for in
1236 * these cases, sharing is handled in the driver or filesystem rather
1237 * than here
1238 */
1239 if (vm_flags & VM_MAYSHARE) {
David Howells8feae132009-01-08 12:04:47 +00001240 struct vm_region *pregion;
1241 unsigned long pglen, rpglen, pgend, rpgend, start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242
David Howells8feae132009-01-08 12:04:47 +00001243 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1244 pgend = pgoff + pglen;
David Howells165b2392007-03-22 00:11:24 -08001245
David Howells8feae132009-01-08 12:04:47 +00001246 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1247 pregion = rb_entry(rb, struct vm_region, vm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248
David Howells8feae132009-01-08 12:04:47 +00001249 if (!(pregion->vm_flags & VM_MAYSHARE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 continue;
1251
1252 /* search for overlapping mappings on the same file */
David Howells8feae132009-01-08 12:04:47 +00001253 if (pregion->vm_file->f_path.dentry->d_inode !=
1254 file->f_path.dentry->d_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 continue;
1256
David Howells8feae132009-01-08 12:04:47 +00001257 if (pregion->vm_pgoff >= pgend)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258 continue;
1259
David Howells8feae132009-01-08 12:04:47 +00001260 rpglen = pregion->vm_end - pregion->vm_start;
1261 rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1262 rpgend = pregion->vm_pgoff + rpglen;
1263 if (pgoff >= rpgend)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 continue;
1265
David Howells8feae132009-01-08 12:04:47 +00001266 /* handle inexactly overlapping matches between
1267 * mappings */
1268 if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1269 !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1270 /* new mapping is not a subset of the region */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 if (!(capabilities & BDI_CAP_MAP_DIRECT))
1272 goto sharing_violation;
1273 continue;
1274 }
1275
David Howells8feae132009-01-08 12:04:47 +00001276 /* we've found a region we can share */
1277 atomic_inc(&pregion->vm_usage);
1278 vma->vm_region = pregion;
1279 start = pregion->vm_start;
1280 start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1281 vma->vm_start = start;
1282 vma->vm_end = start + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283
David Howells8feae132009-01-08 12:04:47 +00001284 if (pregion->vm_flags & VM_MAPPED_COPY) {
1285 kdebug("share copy");
1286 vma->vm_flags |= VM_MAPPED_COPY;
1287 } else {
1288 kdebug("share mmap");
1289 ret = do_mmap_shared_file(vma);
1290 if (ret < 0) {
1291 vma->vm_region = NULL;
1292 vma->vm_start = 0;
1293 vma->vm_end = 0;
1294 atomic_dec(&pregion->vm_usage);
1295 pregion = NULL;
1296 goto error_just_free;
1297 }
1298 }
1299 fput(region->vm_file);
1300 kmem_cache_free(vm_region_jar, region);
1301 region = pregion;
1302 result = start;
1303 goto share;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 }
1305
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 /* obtain the address at which to make a shared mapping
1307 * - this is the hook for quasi-memory character devices to
1308 * tell us the location of a shared mapping
1309 */
1310 if (file && file->f_op->get_unmapped_area) {
1311 addr = file->f_op->get_unmapped_area(file, addr, len,
1312 pgoff, flags);
1313 if (IS_ERR((void *) addr)) {
1314 ret = addr;
1315 if (ret != (unsigned long) -ENOSYS)
David Howells8feae132009-01-08 12:04:47 +00001316 goto error_just_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317
1318 /* the driver refused to tell us where to site
1319 * the mapping so we'll have to attempt to copy
1320 * it */
1321 ret = (unsigned long) -ENODEV;
1322 if (!(capabilities & BDI_CAP_MAP_COPY))
David Howells8feae132009-01-08 12:04:47 +00001323 goto error_just_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324
1325 capabilities &= ~BDI_CAP_MAP_DIRECT;
David Howells8feae132009-01-08 12:04:47 +00001326 } else {
1327 vma->vm_start = region->vm_start = addr;
1328 vma->vm_end = region->vm_end = addr + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329 }
1330 }
1331 }
1332
David Howells8feae132009-01-08 12:04:47 +00001333 vma->vm_region = region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334
1335 /* set up the mapping */
1336 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 Howells8feae132009-01-08 12:04:47 +00001339 ret = do_mmap_private(vma, region, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 if (ret < 0)
David Howells8feae132009-01-08 12:04:47 +00001341 goto error_put_region;
1342
1343 add_nommu_region(region);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344
1345 /* okay... we have a mapping; now we have to register it */
David Howells8feae132009-01-08 12:04:47 +00001346 result = vma->vm_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 current->mm->total_vm += len >> PAGE_SHIFT;
1349
David Howells8feae132009-01-08 12:04:47 +00001350share:
1351 add_vma_to_mm(current->mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352
David Howells8feae132009-01-08 12:04:47 +00001353 up_write(&nommu_region_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354
1355 if (prot & PROT_EXEC)
David Howells8feae132009-01-08 12:04:47 +00001356 flush_icache_range(result, result + len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357
David Howells8feae132009-01-08 12:04:47 +00001358 kleave(" = %lx", result);
1359 return result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360
David Howells8feae132009-01-08 12:04:47 +00001361error_put_region:
1362 __put_nommu_region(region);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 if (vma) {
Matt Helsley925d1c42008-04-29 01:01:36 -07001364 if (vma->vm_file) {
Gavin Lambert3fcd03e2006-09-30 23:27:01 -07001365 fput(vma->vm_file);
Matt Helsley925d1c42008-04-29 01:01:36 -07001366 if (vma->vm_flags & VM_EXECUTABLE)
1367 removed_exe_file_vma(vma->vm_mm);
1368 }
David Howells8feae132009-01-08 12:04:47 +00001369 kmem_cache_free(vm_area_cachep, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 }
David Howells8feae132009-01-08 12:04:47 +00001371 kleave(" = %d [pr]", ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 return ret;
1373
David Howells8feae132009-01-08 12:04:47 +00001374error_just_free:
1375 up_write(&nommu_region_sem);
1376error:
1377 fput(region->vm_file);
1378 kmem_cache_free(vm_region_jar, region);
1379 fput(vma->vm_file);
1380 if (vma->vm_flags & VM_EXECUTABLE)
1381 removed_exe_file_vma(vma->vm_mm);
1382 kmem_cache_free(vm_area_cachep, vma);
1383 kleave(" = %d", ret);
1384 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385
David Howells8feae132009-01-08 12:04:47 +00001386sharing_violation:
1387 up_write(&nommu_region_sem);
1388 printk(KERN_WARNING "Attempt to share mismatched mappings\n");
1389 ret = -EINVAL;
1390 goto error;
1391
1392error_getting_vma:
1393 kmem_cache_free(vm_region_jar, region);
1394 printk(KERN_WARNING "Allocation of vma for %lu byte allocation"
1395 " from process %d failed\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396 len, current->pid);
1397 show_free_areas();
1398 return -ENOMEM;
1399
David Howells8feae132009-01-08 12:04:47 +00001400error_getting_region:
1401 printk(KERN_WARNING "Allocation of vm region for %lu byte allocation"
1402 " from process %d failed\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 len, current->pid);
1404 show_free_areas();
1405 return -ENOMEM;
1406}
Paul Mundtb5073172007-07-21 04:37:25 -07001407EXPORT_SYMBOL(do_mmap_pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408
1409/*
David Howells8feae132009-01-08 12:04:47 +00001410 * split a vma into two pieces at address 'addr', a new vma is allocated either
1411 * for the first part or the tail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 */
David Howells8feae132009-01-08 12:04:47 +00001413int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1414 unsigned long addr, int new_below)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415{
David Howells8feae132009-01-08 12:04:47 +00001416 struct vm_area_struct *new;
1417 struct vm_region *region;
1418 unsigned long npages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419
David Howells8feae132009-01-08 12:04:47 +00001420 kenter("");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421
David Howells8feae132009-01-08 12:04:47 +00001422 /* we're only permitted to split anonymous regions that have a single
1423 * owner */
1424 if (vma->vm_file ||
1425 atomic_read(&vma->vm_region->vm_usage) != 1)
1426 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427
David Howells8feae132009-01-08 12:04:47 +00001428 if (mm->map_count >= sysctl_max_map_count)
1429 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430
David Howells8feae132009-01-08 12:04:47 +00001431 region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1432 if (!region)
1433 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434
David Howells8feae132009-01-08 12:04:47 +00001435 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1436 if (!new) {
1437 kmem_cache_free(vm_region_jar, region);
1438 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439 }
David Howells8feae132009-01-08 12:04:47 +00001440
1441 /* most fields are the same, copy all, and then fixup */
1442 *new = *vma;
1443 *region = *vma->vm_region;
1444 new->vm_region = region;
1445
1446 npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1447
1448 if (new_below) {
Paul Mundtdd8632a2009-01-08 12:04:47 +00001449 region->vm_top = region->vm_end = new->vm_end = addr;
David Howells8feae132009-01-08 12:04:47 +00001450 } else {
1451 region->vm_start = new->vm_start = addr;
1452 region->vm_pgoff = new->vm_pgoff += npages;
1453 }
1454
1455 if (new->vm_ops && new->vm_ops->open)
1456 new->vm_ops->open(new);
1457
1458 delete_vma_from_mm(vma);
1459 down_write(&nommu_region_sem);
1460 delete_nommu_region(vma->vm_region);
1461 if (new_below) {
1462 vma->vm_region->vm_start = vma->vm_start = addr;
1463 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1464 } else {
1465 vma->vm_region->vm_end = vma->vm_end = addr;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001466 vma->vm_region->vm_top = addr;
David Howells8feae132009-01-08 12:04:47 +00001467 }
1468 add_nommu_region(vma->vm_region);
1469 add_nommu_region(new->vm_region);
1470 up_write(&nommu_region_sem);
1471 add_vma_to_mm(mm, vma);
1472 add_vma_to_mm(mm, new);
1473 return 0;
1474}
1475
1476/*
1477 * shrink a VMA by removing the specified chunk from either the beginning or
1478 * the end
1479 */
1480static int shrink_vma(struct mm_struct *mm,
1481 struct vm_area_struct *vma,
1482 unsigned long from, unsigned long to)
1483{
1484 struct vm_region *region;
1485
1486 kenter("");
1487
1488 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1489 * and list */
1490 delete_vma_from_mm(vma);
1491 if (from > vma->vm_start)
1492 vma->vm_end = from;
1493 else
1494 vma->vm_start = to;
1495 add_vma_to_mm(mm, vma);
1496
1497 /* cut the backing region down to size */
1498 region = vma->vm_region;
1499 BUG_ON(atomic_read(&region->vm_usage) != 1);
1500
1501 down_write(&nommu_region_sem);
1502 delete_nommu_region(region);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001503 if (from > region->vm_start) {
1504 to = region->vm_top;
1505 region->vm_top = region->vm_end = from;
1506 } else {
David Howells8feae132009-01-08 12:04:47 +00001507 region->vm_start = to;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001508 }
David Howells8feae132009-01-08 12:04:47 +00001509 add_nommu_region(region);
1510 up_write(&nommu_region_sem);
1511
1512 free_page_series(from, to);
1513 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514}
1515
David Howells30340972006-09-27 01:50:20 -07001516/*
1517 * release a mapping
David Howells8feae132009-01-08 12:04:47 +00001518 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1519 * VMA, though it need not cover the whole VMA
David Howells30340972006-09-27 01:50:20 -07001520 */
David Howells8feae132009-01-08 12:04:47 +00001521int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522{
David Howells8feae132009-01-08 12:04:47 +00001523 struct vm_area_struct *vma;
1524 struct rb_node *rb;
1525 unsigned long end = start + len;
1526 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527
David Howells8feae132009-01-08 12:04:47 +00001528 kenter(",%lx,%zx", start, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529
David Howells8feae132009-01-08 12:04:47 +00001530 if (len == 0)
1531 return -EINVAL;
1532
1533 /* find the first potentially overlapping VMA */
1534 vma = find_vma(mm, start);
1535 if (!vma) {
David Howells33e5d7692009-04-02 16:56:32 -07001536 static int limit = 0;
1537 if (limit < 5) {
1538 printk(KERN_WARNING
1539 "munmap of memory not mmapped by process %d"
1540 " (%s): 0x%lx-0x%lx\n",
1541 current->pid, current->comm,
1542 start, start + len - 1);
1543 limit++;
1544 }
David Howells8feae132009-01-08 12:04:47 +00001545 return -EINVAL;
David Howells30340972006-09-27 01:50:20 -07001546 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547
David Howells8feae132009-01-08 12:04:47 +00001548 /* we're allowed to split an anonymous VMA but not a file-backed one */
1549 if (vma->vm_file) {
1550 do {
1551 if (start > vma->vm_start) {
1552 kleave(" = -EINVAL [miss]");
1553 return -EINVAL;
1554 }
1555 if (end == vma->vm_end)
1556 goto erase_whole_vma;
1557 rb = rb_next(&vma->vm_rb);
1558 vma = rb_entry(rb, struct vm_area_struct, vm_rb);
1559 } while (rb);
1560 kleave(" = -EINVAL [split file]");
1561 return -EINVAL;
1562 } else {
1563 /* the chunk must be a subset of the VMA found */
1564 if (start == vma->vm_start && end == vma->vm_end)
1565 goto erase_whole_vma;
1566 if (start < vma->vm_start || end > vma->vm_end) {
1567 kleave(" = -EINVAL [superset]");
1568 return -EINVAL;
1569 }
1570 if (start & ~PAGE_MASK) {
1571 kleave(" = -EINVAL [unaligned start]");
1572 return -EINVAL;
1573 }
1574 if (end != vma->vm_end && end & ~PAGE_MASK) {
1575 kleave(" = -EINVAL [unaligned split]");
1576 return -EINVAL;
1577 }
1578 if (start != vma->vm_start && end != vma->vm_end) {
1579 ret = split_vma(mm, vma, start, 1);
1580 if (ret < 0) {
1581 kleave(" = %d [split]", ret);
1582 return ret;
1583 }
1584 }
1585 return shrink_vma(mm, vma, start, end);
1586 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587
David Howells8feae132009-01-08 12:04:47 +00001588erase_whole_vma:
1589 delete_vma_from_mm(vma);
1590 delete_vma(mm, vma);
1591 kleave(" = 0");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592 return 0;
1593}
Paul Mundtb5073172007-07-21 04:37:25 -07001594EXPORT_SYMBOL(do_munmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595
Heiko Carstens6a6160a2009-01-14 14:14:15 +01001596SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
David Howells30340972006-09-27 01:50:20 -07001597{
1598 int ret;
1599 struct mm_struct *mm = current->mm;
1600
1601 down_write(&mm->mmap_sem);
1602 ret = do_munmap(mm, addr, len);
1603 up_write(&mm->mmap_sem);
1604 return ret;
1605}
1606
1607/*
David Howells8feae132009-01-08 12:04:47 +00001608 * release all the mappings made in a process's VM space
David Howells30340972006-09-27 01:50:20 -07001609 */
David Howells8feae132009-01-08 12:04:47 +00001610void exit_mmap(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611{
David Howells8feae132009-01-08 12:04:47 +00001612 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613
David Howells8feae132009-01-08 12:04:47 +00001614 if (!mm)
1615 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616
David Howells8feae132009-01-08 12:04:47 +00001617 kenter("");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618
David Howells8feae132009-01-08 12:04:47 +00001619 mm->total_vm = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620
David Howells8feae132009-01-08 12:04:47 +00001621 while ((vma = mm->mmap)) {
1622 mm->mmap = vma->vm_next;
1623 delete_vma_from_mm(vma);
1624 delete_vma(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625 }
David Howells8feae132009-01-08 12:04:47 +00001626
1627 kleave("");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628}
1629
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630unsigned long do_brk(unsigned long addr, unsigned long len)
1631{
1632 return -ENOMEM;
1633}
1634
1635/*
David Howells6fa5f802006-09-27 01:50:21 -07001636 * expand (or shrink) an existing mapping, potentially moving it at the same
1637 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638 *
David Howells6fa5f802006-09-27 01:50:21 -07001639 * under NOMMU conditions, we only permit changing a mapping's size, and only
David Howells8feae132009-01-08 12:04:47 +00001640 * as long as it stays within the region allocated by do_mmap_private() and the
1641 * block is not shareable
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642 *
David Howells6fa5f802006-09-27 01:50:21 -07001643 * MREMAP_FIXED is not supported under NOMMU conditions
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 */
1645unsigned long do_mremap(unsigned long addr,
1646 unsigned long old_len, unsigned long new_len,
1647 unsigned long flags, unsigned long new_addr)
1648{
David Howells6fa5f802006-09-27 01:50:21 -07001649 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650
1651 /* insanity checks first */
David Howells8feae132009-01-08 12:04:47 +00001652 if (old_len == 0 || new_len == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 return (unsigned long) -EINVAL;
1654
David Howells8feae132009-01-08 12:04:47 +00001655 if (addr & ~PAGE_MASK)
1656 return -EINVAL;
1657
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 if (flags & MREMAP_FIXED && new_addr != addr)
1659 return (unsigned long) -EINVAL;
1660
David Howells8feae132009-01-08 12:04:47 +00001661 vma = find_vma_exact(current->mm, addr, old_len);
David Howells6fa5f802006-09-27 01:50:21 -07001662 if (!vma)
1663 return (unsigned long) -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664
David Howells6fa5f802006-09-27 01:50:21 -07001665 if (vma->vm_end != vma->vm_start + old_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666 return (unsigned long) -EFAULT;
1667
David Howells6fa5f802006-09-27 01:50:21 -07001668 if (vma->vm_flags & VM_MAYSHARE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669 return (unsigned long) -EPERM;
1670
David Howells8feae132009-01-08 12:04:47 +00001671 if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672 return (unsigned long) -ENOMEM;
1673
1674 /* all checks complete - do it */
David Howells6fa5f802006-09-27 01:50:21 -07001675 vma->vm_end = vma->vm_start + new_len;
David Howells6fa5f802006-09-27 01:50:21 -07001676 return vma->vm_start;
1677}
Paul Mundtb5073172007-07-21 04:37:25 -07001678EXPORT_SYMBOL(do_mremap);
David Howells6fa5f802006-09-27 01:50:21 -07001679
Heiko Carstens6a6160a2009-01-14 14:14:15 +01001680SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1681 unsigned long, new_len, unsigned long, flags,
1682 unsigned long, new_addr)
David Howells6fa5f802006-09-27 01:50:21 -07001683{
1684 unsigned long ret;
1685
1686 down_write(&current->mm->mmap_sem);
1687 ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1688 up_write(&current->mm->mmap_sem);
1689 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690}
1691
Linus Torvalds6aab3412005-11-28 14:34:23 -08001692struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001693 unsigned int foll_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694{
1695 return NULL;
1696}
1697
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698int remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
1699 unsigned long to, unsigned long size, pgprot_t prot)
1700{
Greg Ungerer66aa2b42005-09-12 11:18:10 +10001701 vma->vm_start = vma->vm_pgoff << PAGE_SHIFT;
1702 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703}
Luke Yang22c4af42006-07-14 00:24:09 -07001704EXPORT_SYMBOL(remap_pfn_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705
Paul Mundtf905bc42008-02-04 22:29:59 -08001706int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1707 unsigned long pgoff)
1708{
1709 unsigned int size = vma->vm_end - vma->vm_start;
1710
1711 if (!(vma->vm_flags & VM_USERMAP))
1712 return -EINVAL;
1713
1714 vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1715 vma->vm_end = vma->vm_start + size;
1716
1717 return 0;
1718}
1719EXPORT_SYMBOL(remap_vmalloc_range);
1720
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1722{
1723}
1724
1725unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1726 unsigned long len, unsigned long pgoff, unsigned long flags)
1727{
1728 return -ENOMEM;
1729}
1730
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001731void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732{
1733}
1734
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735void unmap_mapping_range(struct address_space *mapping,
1736 loff_t const holebegin, loff_t const holelen,
1737 int even_cows)
1738{
1739}
Luke Yang22c4af42006-07-14 00:24:09 -07001740EXPORT_SYMBOL(unmap_mapping_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741
1742/*
David Howellsd56e03c2007-03-22 00:11:23 -08001743 * ask for an unmapped area at which to create a mapping on a file
1744 */
1745unsigned long get_unmapped_area(struct file *file, unsigned long addr,
1746 unsigned long len, unsigned long pgoff,
1747 unsigned long flags)
1748{
1749 unsigned long (*get_area)(struct file *, unsigned long, unsigned long,
1750 unsigned long, unsigned long);
1751
1752 get_area = current->mm->get_unmapped_area;
1753 if (file && file->f_op && file->f_op->get_unmapped_area)
1754 get_area = file->f_op->get_unmapped_area;
1755
1756 if (!get_area)
1757 return -ENOSYS;
1758
1759 return get_area(file, addr, len, pgoff, flags);
1760}
David Howellsd56e03c2007-03-22 00:11:23 -08001761EXPORT_SYMBOL(get_unmapped_area);
1762
1763/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764 * Check that a process has enough memory to allocate a new virtual
1765 * mapping. 0 means there is enough memory for the allocation to
1766 * succeed and -ENOMEM implies there is not.
1767 *
1768 * We currently support three overcommit policies, which are set via the
1769 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
1770 *
1771 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1772 * Additional code 2002 Jul 20 by Robert Love.
1773 *
1774 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1775 *
1776 * Note this is a helper function intended to be used by LSMs which
1777 * wish to use this logic.
1778 */
Alan Cox34b4e4a2007-08-22 14:01:28 -07001779int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780{
1781 unsigned long free, allowed;
1782
1783 vm_acct_memory(pages);
1784
1785 /*
1786 * Sometimes we want to use more memory than we have
1787 */
1788 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1789 return 0;
1790
1791 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1792 unsigned long n;
1793
Christoph Lameter347ce432006-06-30 01:55:35 -07001794 free = global_page_state(NR_FILE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795 free += nr_swap_pages;
1796
1797 /*
1798 * Any slabs which are created with the
1799 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1800 * which are reclaimable, under pressure. The dentry
1801 * cache and most inode caches should fall into this
1802 */
Christoph Lameter972d1a72006-09-25 23:31:51 -07001803 free += global_page_state(NR_SLAB_RECLAIMABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804
1805 /*
1806 * Leave the last 3% for root
1807 */
1808 if (!cap_sys_admin)
1809 free -= free / 32;
1810
1811 if (free > pages)
1812 return 0;
1813
1814 /*
1815 * nr_free_pages() is very expensive on large systems,
1816 * only call if we're about to fail.
1817 */
1818 n = nr_free_pages();
Hideo AOKId5ddc792006-04-10 22:53:01 -07001819
1820 /*
1821 * Leave reserved pages. The pages are not for anonymous pages.
1822 */
1823 if (n <= totalreserve_pages)
1824 goto error;
1825 else
1826 n -= totalreserve_pages;
1827
1828 /*
1829 * Leave the last 3% for root
1830 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831 if (!cap_sys_admin)
1832 n -= n / 32;
1833 free += n;
1834
1835 if (free > pages)
1836 return 0;
Hideo AOKId5ddc792006-04-10 22:53:01 -07001837
1838 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839 }
1840
1841 allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1842 /*
1843 * Leave the last 3% for root
1844 */
1845 if (!cap_sys_admin)
1846 allowed -= allowed / 32;
1847 allowed += total_swap_pages;
1848
1849 /* Don't let a single process grow too big:
1850 leave 3% of the size of this process for other processes */
Alan Cox731572d2008-10-29 14:01:20 -07001851 if (mm)
1852 allowed -= mm->total_vm / 32;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07001854 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855 return 0;
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07001856
Hideo AOKId5ddc792006-04-10 22:53:01 -07001857error:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858 vm_unacct_memory(pages);
1859
1860 return -ENOMEM;
1861}
1862
1863int in_gate_area_no_task(unsigned long addr)
1864{
1865 return 0;
1866}
David Howellsb0e15192006-01-06 00:11:42 -08001867
Nick Piggind0217ac2007-07-19 01:47:03 -07001868int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
David Howellsb0e15192006-01-06 00:11:42 -08001869{
1870 BUG();
Nick Piggind0217ac2007-07-19 01:47:03 -07001871 return 0;
David Howellsb0e15192006-01-06 00:11:42 -08001872}
Paul Mundtb5073172007-07-21 04:37:25 -07001873EXPORT_SYMBOL(filemap_fault);
David Howells0ec76a12006-09-27 01:50:15 -07001874
1875/*
1876 * Access another process' address space.
1877 * - source/target buffer must be kernel space
1878 */
1879int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
1880{
David Howells0ec76a12006-09-27 01:50:15 -07001881 struct vm_area_struct *vma;
1882 struct mm_struct *mm;
1883
1884 if (addr + len < addr)
1885 return 0;
1886
1887 mm = get_task_mm(tsk);
1888 if (!mm)
1889 return 0;
1890
1891 down_read(&mm->mmap_sem);
1892
1893 /* the access must start within one of the target process's mappings */
David Howells0159b142006-09-27 01:50:16 -07001894 vma = find_vma(mm, addr);
1895 if (vma) {
David Howells0ec76a12006-09-27 01:50:15 -07001896 /* don't overrun this mapping */
1897 if (addr + len >= vma->vm_end)
1898 len = vma->vm_end - addr;
1899
1900 /* only read or write mappings where it is permitted */
David Howellsd00c7b992006-09-27 01:50:19 -07001901 if (write && vma->vm_flags & VM_MAYWRITE)
David Howells0ec76a12006-09-27 01:50:15 -07001902 len -= copy_to_user((void *) addr, buf, len);
David Howellsd00c7b992006-09-27 01:50:19 -07001903 else if (!write && vma->vm_flags & VM_MAYREAD)
David Howells0ec76a12006-09-27 01:50:15 -07001904 len -= copy_from_user(buf, (void *) addr, len);
1905 else
1906 len = 0;
1907 } else {
1908 len = 0;
1909 }
1910
1911 up_read(&mm->mmap_sem);
1912 mmput(mm);
1913 return len;
1914}