blob: b571ef707428c5e171fa27b7fc09d0212018c5df [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;
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
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 Mundt6cfd53f2008-06-05 22:46:08 -0700140 return 0;
141
142 page = virt_to_head_page(objp);
Paul Mundt6cfd53f2008-06-05 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 Mundt6cfd53f2008-06-05 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 Mundt6cfd53f2008-06-05 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 Mundt6cfd53f2008-06-05 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
David Howells8feae132009-01-08 12:04:47 +0000518 parent = NULL;
519 p = &nommu_region_tree.rb_node;
520 while (*p) {
521 parent = *p;
522 pregion = rb_entry(parent, struct vm_region, vm_rb);
523 if (region->vm_start < pregion->vm_start)
524 p = &(*p)->rb_left;
525 else if (region->vm_start > pregion->vm_start)
526 p = &(*p)->rb_right;
527 else if (pregion == region)
528 return;
529 else
530 BUG();
531 }
532
533 rb_link_node(&region->vm_rb, parent, p);
534 rb_insert_color(&region->vm_rb, &nommu_region_tree);
535
536 validate_nommu_regions();
537}
538
539/*
540 * delete a region from the global tree
541 */
542static void delete_nommu_region(struct vm_region *region)
543{
544 BUG_ON(!nommu_region_tree.rb_node);
545
546 validate_nommu_regions();
547 rb_erase(&region->vm_rb, &nommu_region_tree);
548 validate_nommu_regions();
549}
550
551/*
552 * free a contiguous series of pages
553 */
554static void free_page_series(unsigned long from, unsigned long to)
555{
556 for (; from < to; from += PAGE_SIZE) {
557 struct page *page = virt_to_page(from);
558
559 kdebug("- free %lx", from);
David Howells33e5d7692009-04-02 16:56:32 -0700560 atomic_long_dec(&mmap_pages_allocated);
David Howells8feae132009-01-08 12:04:47 +0000561 if (page_count(page) != 1)
David Howells33e5d7692009-04-02 16:56:32 -0700562 kdebug("free page %p: refcount not one: %d",
563 page, page_count(page));
David Howells8feae132009-01-08 12:04:47 +0000564 put_page(page);
565 }
566}
567
568/*
569 * release a reference to a region
David Howells33e5d7692009-04-02 16:56:32 -0700570 * - the caller must hold the region semaphore for writing, which this releases
Paul Mundtdd8632a2009-01-08 12:04:47 +0000571 * - the region may not have been added to the tree yet, in which case vm_top
David Howells8feae132009-01-08 12:04:47 +0000572 * will equal vm_start
573 */
574static void __put_nommu_region(struct vm_region *region)
575 __releases(nommu_region_sem)
576{
577 kenter("%p{%d}", region, atomic_read(&region->vm_usage));
578
579 BUG_ON(!nommu_region_tree.rb_node);
580
581 if (atomic_dec_and_test(&region->vm_usage)) {
Paul Mundtdd8632a2009-01-08 12:04:47 +0000582 if (region->vm_top > region->vm_start)
David Howells8feae132009-01-08 12:04:47 +0000583 delete_nommu_region(region);
584 up_write(&nommu_region_sem);
585
586 if (region->vm_file)
587 fput(region->vm_file);
588
589 /* IO memory and memory shared directly out of the pagecache
590 * from ramfs/tmpfs mustn't be released here */
591 if (region->vm_flags & VM_MAPPED_COPY) {
592 kdebug("free series");
Paul Mundtdd8632a2009-01-08 12:04:47 +0000593 free_page_series(region->vm_start, region->vm_top);
David Howells8feae132009-01-08 12:04:47 +0000594 }
595 kmem_cache_free(vm_region_jar, region);
596 } else {
597 up_write(&nommu_region_sem);
598 }
599}
600
601/*
602 * release a reference to a region
603 */
604static void put_nommu_region(struct vm_region *region)
605{
606 down_write(&nommu_region_sem);
607 __put_nommu_region(region);
608}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609
David Howells30340972006-09-27 01:50:20 -0700610/*
611 * add a VMA into a process's mm_struct in the appropriate place in the list
David Howells8feae132009-01-08 12:04:47 +0000612 * and tree and add to the address space's page tree also if not an anonymous
613 * page
David Howells30340972006-09-27 01:50:20 -0700614 * - should be called with mm->mmap_sem held writelocked
615 */
David Howells8feae132009-01-08 12:04:47 +0000616static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
David Howells30340972006-09-27 01:50:20 -0700617{
David Howells8feae132009-01-08 12:04:47 +0000618 struct vm_area_struct *pvma, **pp;
619 struct address_space *mapping;
620 struct rb_node **p, *parent;
David Howells30340972006-09-27 01:50:20 -0700621
David Howells8feae132009-01-08 12:04:47 +0000622 kenter(",%p", vma);
623
624 BUG_ON(!vma->vm_region);
625
626 mm->map_count++;
627 vma->vm_mm = mm;
628
629 /* add the VMA to the mapping */
630 if (vma->vm_file) {
631 mapping = vma->vm_file->f_mapping;
632
633 flush_dcache_mmap_lock(mapping);
634 vma_prio_tree_insert(vma, &mapping->i_mmap);
635 flush_dcache_mmap_unlock(mapping);
636 }
637
638 /* add the VMA to the tree */
639 parent = NULL;
640 p = &mm->mm_rb.rb_node;
641 while (*p) {
642 parent = *p;
643 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
644
645 /* sort by: start addr, end addr, VMA struct addr in that order
646 * (the latter is necessary as we may get identical VMAs) */
647 if (vma->vm_start < pvma->vm_start)
648 p = &(*p)->rb_left;
649 else if (vma->vm_start > pvma->vm_start)
650 p = &(*p)->rb_right;
651 else if (vma->vm_end < pvma->vm_end)
652 p = &(*p)->rb_left;
653 else if (vma->vm_end > pvma->vm_end)
654 p = &(*p)->rb_right;
655 else if (vma < pvma)
656 p = &(*p)->rb_left;
657 else if (vma > pvma)
658 p = &(*p)->rb_right;
659 else
660 BUG();
661 }
662
663 rb_link_node(&vma->vm_rb, parent, p);
664 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
665
666 /* add VMA to the VMA list also */
667 for (pp = &mm->mmap; (pvma = *pp); pp = &(*pp)->vm_next) {
668 if (pvma->vm_start > vma->vm_start)
David Howells30340972006-09-27 01:50:20 -0700669 break;
David Howells8feae132009-01-08 12:04:47 +0000670 if (pvma->vm_start < vma->vm_start)
671 continue;
672 if (pvma->vm_end < vma->vm_end)
673 break;
674 }
David Howells30340972006-09-27 01:50:20 -0700675
David Howells8feae132009-01-08 12:04:47 +0000676 vma->vm_next = *pp;
677 *pp = vma;
678}
679
680/*
681 * delete a VMA from its owning mm_struct and address space
682 */
683static void delete_vma_from_mm(struct vm_area_struct *vma)
684{
685 struct vm_area_struct **pp;
686 struct address_space *mapping;
687 struct mm_struct *mm = vma->vm_mm;
688
689 kenter("%p", vma);
690
691 mm->map_count--;
692 if (mm->mmap_cache == vma)
693 mm->mmap_cache = NULL;
694
695 /* remove the VMA from the mapping */
696 if (vma->vm_file) {
697 mapping = vma->vm_file->f_mapping;
698
699 flush_dcache_mmap_lock(mapping);
700 vma_prio_tree_remove(vma, &mapping->i_mmap);
701 flush_dcache_mmap_unlock(mapping);
702 }
703
704 /* remove from the MM's tree and list */
705 rb_erase(&vma->vm_rb, &mm->mm_rb);
706 for (pp = &mm->mmap; *pp; pp = &(*pp)->vm_next) {
707 if (*pp == vma) {
708 *pp = vma->vm_next;
709 break;
710 }
711 }
712
713 vma->vm_mm = NULL;
714}
715
716/*
717 * destroy a VMA record
718 */
719static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
720{
721 kenter("%p", vma);
722 if (vma->vm_ops && vma->vm_ops->close)
723 vma->vm_ops->close(vma);
724 if (vma->vm_file) {
725 fput(vma->vm_file);
726 if (vma->vm_flags & VM_EXECUTABLE)
727 removed_exe_file_vma(mm);
728 }
729 put_nommu_region(vma->vm_region);
730 kmem_cache_free(vm_area_cachep, vma);
David Howells30340972006-09-27 01:50:20 -0700731}
732
733/*
734 * look up the first VMA in which addr resides, NULL if none
735 * - should be called with mm->mmap_sem at least held readlocked
736 */
737struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
738{
David Howells8feae132009-01-08 12:04:47 +0000739 struct vm_area_struct *vma;
740 struct rb_node *n = mm->mm_rb.rb_node;
David Howells30340972006-09-27 01:50:20 -0700741
David Howells8feae132009-01-08 12:04:47 +0000742 /* check the cache first */
743 vma = mm->mmap_cache;
744 if (vma && vma->vm_start <= addr && vma->vm_end > addr)
745 return vma;
746
747 /* trawl the tree (there may be multiple mappings in which addr
748 * resides) */
749 for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) {
750 vma = rb_entry(n, struct vm_area_struct, vm_rb);
751 if (vma->vm_start > addr)
752 return NULL;
753 if (vma->vm_end > addr) {
754 mm->mmap_cache = vma;
755 return vma;
756 }
David Howells30340972006-09-27 01:50:20 -0700757 }
758
David Howells30340972006-09-27 01:50:20 -0700759 return NULL;
760}
761EXPORT_SYMBOL(find_vma);
762
763/*
David Howells930e6522006-09-27 01:50:22 -0700764 * find a VMA
765 * - we don't extend stack VMAs under NOMMU conditions
766 */
767struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
768{
769 return find_vma(mm, addr);
770}
771
David Howells8feae132009-01-08 12:04:47 +0000772/*
773 * expand a stack to a given address
774 * - not supported under NOMMU conditions
775 */
Greg Ungerer57c8f632007-07-15 23:38:28 -0700776int expand_stack(struct vm_area_struct *vma, unsigned long address)
777{
778 return -ENOMEM;
779}
780
David Howells930e6522006-09-27 01:50:22 -0700781/*
David Howells6fa5f802006-09-27 01:50:21 -0700782 * look up the first VMA exactly that exactly matches addr
783 * - should be called with mm->mmap_sem at least held readlocked
784 */
David Howells8feae132009-01-08 12:04:47 +0000785static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
786 unsigned long addr,
787 unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788{
789 struct vm_area_struct *vma;
David Howells8feae132009-01-08 12:04:47 +0000790 struct rb_node *n = mm->mm_rb.rb_node;
791 unsigned long end = addr + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792
David Howells8feae132009-01-08 12:04:47 +0000793 /* check the cache first */
794 vma = mm->mmap_cache;
795 if (vma && vma->vm_start == addr && vma->vm_end == end)
796 return vma;
797
798 /* trawl the tree (there may be multiple mappings in which addr
799 * resides) */
800 for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 vma = rb_entry(n, struct vm_area_struct, vm_rb);
David Howells8feae132009-01-08 12:04:47 +0000802 if (vma->vm_start < addr)
803 continue;
804 if (vma->vm_start > addr)
805 return NULL;
806 if (vma->vm_end == end) {
807 mm->mmap_cache = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 return vma;
David Howells8feae132009-01-08 12:04:47 +0000809 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 }
811
812 return NULL;
813}
814
David Howells30340972006-09-27 01:50:20 -0700815/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 * determine whether a mapping should be permitted and, if so, what sort of
817 * mapping we're capable of supporting
818 */
819static int validate_mmap_request(struct file *file,
820 unsigned long addr,
821 unsigned long len,
822 unsigned long prot,
823 unsigned long flags,
824 unsigned long pgoff,
825 unsigned long *_capabilities)
826{
David Howells8feae132009-01-08 12:04:47 +0000827 unsigned long capabilities, rlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 unsigned long reqprot = prot;
829 int ret;
830
831 /* do the simple checks first */
832 if (flags & MAP_FIXED || addr) {
833 printk(KERN_DEBUG
834 "%d: Can't do fixed-address/overlay mmap of RAM\n",
835 current->pid);
836 return -EINVAL;
837 }
838
839 if ((flags & MAP_TYPE) != MAP_PRIVATE &&
840 (flags & MAP_TYPE) != MAP_SHARED)
841 return -EINVAL;
842
Mike Frysingerf81cff02006-12-06 12:02:59 +1000843 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 return -EINVAL;
845
Mike Frysingerf81cff02006-12-06 12:02:59 +1000846 /* Careful about overflows.. */
David Howells8feae132009-01-08 12:04:47 +0000847 rlen = PAGE_ALIGN(len);
848 if (!rlen || rlen > TASK_SIZE)
Mike Frysingerf81cff02006-12-06 12:02:59 +1000849 return -ENOMEM;
850
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 /* offset overflow? */
David Howells8feae132009-01-08 12:04:47 +0000852 if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
Mike Frysingerf81cff02006-12-06 12:02:59 +1000853 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854
855 if (file) {
856 /* validate file mapping requests */
857 struct address_space *mapping;
858
859 /* files must support mmap */
860 if (!file->f_op || !file->f_op->mmap)
861 return -ENODEV;
862
863 /* work out if what we've got could possibly be shared
864 * - we support chardevs that provide their own "memory"
865 * - we support files/blockdevs that are memory backed
866 */
867 mapping = file->f_mapping;
868 if (!mapping)
Josef Sipeke9536ae2006-12-08 02:37:21 -0800869 mapping = file->f_path.dentry->d_inode->i_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870
871 capabilities = 0;
872 if (mapping && mapping->backing_dev_info)
873 capabilities = mapping->backing_dev_info->capabilities;
874
875 if (!capabilities) {
876 /* no explicit capabilities set, so assume some
877 * defaults */
Josef Sipeke9536ae2006-12-08 02:37:21 -0800878 switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 case S_IFREG:
880 case S_IFBLK:
881 capabilities = BDI_CAP_MAP_COPY;
882 break;
883
884 case S_IFCHR:
885 capabilities =
886 BDI_CAP_MAP_DIRECT |
887 BDI_CAP_READ_MAP |
888 BDI_CAP_WRITE_MAP;
889 break;
890
891 default:
892 return -EINVAL;
893 }
894 }
895
896 /* eliminate any capabilities that we can't support on this
897 * device */
898 if (!file->f_op->get_unmapped_area)
899 capabilities &= ~BDI_CAP_MAP_DIRECT;
900 if (!file->f_op->read)
901 capabilities &= ~BDI_CAP_MAP_COPY;
902
903 if (flags & MAP_SHARED) {
904 /* do checks for writing, appending and locking */
905 if ((prot & PROT_WRITE) &&
906 !(file->f_mode & FMODE_WRITE))
907 return -EACCES;
908
Josef Sipeke9536ae2006-12-08 02:37:21 -0800909 if (IS_APPEND(file->f_path.dentry->d_inode) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 (file->f_mode & FMODE_WRITE))
911 return -EACCES;
912
Josef Sipeke9536ae2006-12-08 02:37:21 -0800913 if (locks_verify_locked(file->f_path.dentry->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 return -EAGAIN;
915
916 if (!(capabilities & BDI_CAP_MAP_DIRECT))
917 return -ENODEV;
918
919 if (((prot & PROT_READ) && !(capabilities & BDI_CAP_READ_MAP)) ||
920 ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
921 ((prot & PROT_EXEC) && !(capabilities & BDI_CAP_EXEC_MAP))
922 ) {
923 printk("MAP_SHARED not completely supported on !MMU\n");
924 return -EINVAL;
925 }
926
927 /* we mustn't privatise shared mappings */
928 capabilities &= ~BDI_CAP_MAP_COPY;
929 }
930 else {
931 /* we're going to read the file into private memory we
932 * allocate */
933 if (!(capabilities & BDI_CAP_MAP_COPY))
934 return -ENODEV;
935
936 /* we don't permit a private writable mapping to be
937 * shared with the backing device */
938 if (prot & PROT_WRITE)
939 capabilities &= ~BDI_CAP_MAP_DIRECT;
940 }
941
942 /* handle executable mappings and implied executable
943 * mappings */
Josef Sipeke9536ae2006-12-08 02:37:21 -0800944 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 if (prot & PROT_EXEC)
946 return -EPERM;
947 }
948 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
949 /* handle implication of PROT_EXEC by PROT_READ */
950 if (current->personality & READ_IMPLIES_EXEC) {
951 if (capabilities & BDI_CAP_EXEC_MAP)
952 prot |= PROT_EXEC;
953 }
954 }
955 else if ((prot & PROT_READ) &&
956 (prot & PROT_EXEC) &&
957 !(capabilities & BDI_CAP_EXEC_MAP)
958 ) {
959 /* backing file is not executable, try to copy */
960 capabilities &= ~BDI_CAP_MAP_DIRECT;
961 }
962 }
963 else {
964 /* anonymous mappings are always memory backed and can be
965 * privately mapped
966 */
967 capabilities = BDI_CAP_MAP_COPY;
968
969 /* handle PROT_EXEC implication by PROT_READ */
970 if ((prot & PROT_READ) &&
971 (current->personality & READ_IMPLIES_EXEC))
972 prot |= PROT_EXEC;
973 }
974
975 /* allow the security API to have its say */
Eric Parised032182007-06-28 15:55:21 -0400976 ret = security_file_mmap(file, reqprot, prot, flags, addr, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 if (ret < 0)
978 return ret;
979
980 /* looks okay */
981 *_capabilities = capabilities;
982 return 0;
983}
984
985/*
986 * we've determined that we can make the mapping, now translate what we
987 * now know into VMA flags
988 */
989static unsigned long determine_vm_flags(struct file *file,
990 unsigned long prot,
991 unsigned long flags,
992 unsigned long capabilities)
993{
994 unsigned long vm_flags;
995
996 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
997 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
998 /* vm_flags |= mm->def_flags; */
999
1000 if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
1001 /* attempt to share read-only copies of mapped file chunks */
1002 if (file && !(prot & PROT_WRITE))
1003 vm_flags |= VM_MAYSHARE;
1004 }
1005 else {
1006 /* overlay a shareable mapping on the backing device or inode
1007 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1008 * romfs/cramfs */
1009 if (flags & MAP_SHARED)
1010 vm_flags |= VM_MAYSHARE | VM_SHARED;
1011 else if ((((vm_flags & capabilities) ^ vm_flags) & BDI_CAP_VMFLAGS) == 0)
1012 vm_flags |= VM_MAYSHARE;
1013 }
1014
1015 /* refuse to let anyone share private mappings with this process if
1016 * it's being traced - otherwise breakpoints set in it may interfere
1017 * with another untraced process
1018 */
Roland McGrathfa8e26c2008-07-25 19:45:50 -07001019 if ((flags & MAP_PRIVATE) && tracehook_expect_breakpoints(current))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 vm_flags &= ~VM_MAYSHARE;
1021
1022 return vm_flags;
1023}
1024
1025/*
David Howells8feae132009-01-08 12:04:47 +00001026 * set up a shared mapping on a file (the driver or filesystem provides and
1027 * pins the storage)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 */
David Howells8feae132009-01-08 12:04:47 +00001029static int do_mmap_shared_file(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030{
1031 int ret;
1032
1033 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001034 if (ret == 0) {
1035 vma->vm_region->vm_top = vma->vm_region->vm_end;
1036 return ret;
1037 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 if (ret != -ENOSYS)
1039 return ret;
1040
1041 /* getting an ENOSYS error indicates that direct mmap isn't
1042 * possible (as opposed to tried but failed) so we'll fall
1043 * through to making a private copy of the data and mapping
1044 * that if we can */
1045 return -ENODEV;
1046}
1047
1048/*
1049 * set up a private mapping or an anonymous shared mapping
1050 */
David Howells8feae132009-01-08 12:04:47 +00001051static int do_mmap_private(struct vm_area_struct *vma,
1052 struct vm_region *region,
1053 unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054{
David Howells8feae132009-01-08 12:04:47 +00001055 struct page *pages;
1056 unsigned long total, point, n, rlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 void *base;
David Howells8feae132009-01-08 12:04:47 +00001058 int ret, order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059
1060 /* invoke the file's mapping function so that it can keep track of
1061 * shared mappings on devices or memory
1062 * - VM_MAYSHARE will be set if it may attempt to share
1063 */
1064 if (vma->vm_file) {
1065 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001066 if (ret == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 /* shouldn't return success if we're not sharing */
Paul Mundtdd8632a2009-01-08 12:04:47 +00001068 BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1069 vma->vm_region->vm_top = vma->vm_region->vm_end;
1070 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071 }
Paul Mundtdd8632a2009-01-08 12:04:47 +00001072 if (ret != -ENOSYS)
1073 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074
1075 /* getting an ENOSYS error indicates that direct mmap isn't
1076 * possible (as opposed to tried but failed) so we'll try to
1077 * make a private copy of the data and map that instead */
1078 }
1079
David Howells8feae132009-01-08 12:04:47 +00001080 rlen = PAGE_ALIGN(len);
1081
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 /* allocate some memory to hold the mapping
1083 * - note that this may not return a page-aligned address if the object
1084 * we're allocating is smaller than a page
1085 */
David Howells8feae132009-01-08 12:04:47 +00001086 order = get_order(rlen);
1087 kdebug("alloc order %d for %lx", order, len);
1088
1089 pages = alloc_pages(GFP_KERNEL, order);
1090 if (!pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 goto enomem;
1092
David Howells8feae132009-01-08 12:04:47 +00001093 total = 1 << order;
David Howells33e5d7692009-04-02 16:56:32 -07001094 atomic_long_add(total, &mmap_pages_allocated);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095
David Howells8feae132009-01-08 12:04:47 +00001096 point = rlen >> PAGE_SHIFT;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001097
1098 /* we allocated a power-of-2 sized page set, so we may want to trim off
1099 * the excess */
1100 if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) {
1101 while (total > point) {
1102 order = ilog2(total - point);
1103 n = 1 << order;
1104 kdebug("shave %lu/%lu @%lu", n, total - point, total);
David Howells33e5d7692009-04-02 16:56:32 -07001105 atomic_long_sub(n, &mmap_pages_allocated);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001106 total -= n;
1107 set_page_refcounted(pages + total);
1108 __free_pages(pages + total, order);
1109 }
David Howells8feae132009-01-08 12:04:47 +00001110 }
1111
David Howells8feae132009-01-08 12:04:47 +00001112 for (point = 1; point < total; point++)
1113 set_page_refcounted(&pages[point]);
1114
1115 base = page_address(pages);
1116 region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1117 region->vm_start = (unsigned long) base;
1118 region->vm_end = region->vm_start + rlen;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001119 region->vm_top = region->vm_start + (total << PAGE_SHIFT);
David Howells8feae132009-01-08 12:04:47 +00001120
1121 vma->vm_start = region->vm_start;
1122 vma->vm_end = region->vm_start + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123
1124 if (vma->vm_file) {
1125 /* read the contents of a file into the copy */
1126 mm_segment_t old_fs;
1127 loff_t fpos;
1128
1129 fpos = vma->vm_pgoff;
1130 fpos <<= PAGE_SHIFT;
1131
1132 old_fs = get_fs();
1133 set_fs(KERNEL_DS);
David Howells8feae132009-01-08 12:04:47 +00001134 ret = vma->vm_file->f_op->read(vma->vm_file, base, rlen, &fpos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 set_fs(old_fs);
1136
1137 if (ret < 0)
1138 goto error_free;
1139
1140 /* clear the last little bit */
David Howells8feae132009-01-08 12:04:47 +00001141 if (ret < rlen)
1142 memset(base + ret, 0, rlen - ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143
1144 } else {
1145 /* if it's an anonymous mapping, then just clear it */
David Howells8feae132009-01-08 12:04:47 +00001146 memset(base, 0, rlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 }
1148
1149 return 0;
1150
1151error_free:
David Howells8feae132009-01-08 12:04:47 +00001152 free_page_series(region->vm_start, region->vm_end);
1153 region->vm_start = vma->vm_start = 0;
1154 region->vm_end = vma->vm_end = 0;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001155 region->vm_top = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 return ret;
1157
1158enomem:
Greg Ungerer05ae6fa2009-01-13 17:30:22 +10001159 printk("Allocation of length %lu from process %d (%s) failed\n",
1160 len, current->pid, current->comm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 show_free_areas();
1162 return -ENOMEM;
1163}
1164
1165/*
1166 * handle mapping creation for uClinux
1167 */
1168unsigned long do_mmap_pgoff(struct file *file,
1169 unsigned long addr,
1170 unsigned long len,
1171 unsigned long prot,
1172 unsigned long flags,
1173 unsigned long pgoff)
1174{
David Howells8feae132009-01-08 12:04:47 +00001175 struct vm_area_struct *vma;
1176 struct vm_region *region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 struct rb_node *rb;
David Howells8feae132009-01-08 12:04:47 +00001178 unsigned long capabilities, vm_flags, result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 int ret;
1180
David Howells8feae132009-01-08 12:04:47 +00001181 kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff);
1182
Eric Paris7cd94142007-11-26 18:47:40 -05001183 if (!(flags & MAP_FIXED))
1184 addr = round_hint_to_min(addr);
1185
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 /* decide whether we should attempt the mapping, and if so what sort of
1187 * mapping */
1188 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1189 &capabilities);
David Howells8feae132009-01-08 12:04:47 +00001190 if (ret < 0) {
1191 kleave(" = %d [val]", ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 return ret;
David Howells8feae132009-01-08 12:04:47 +00001193 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194
1195 /* we've determined that we can make the mapping, now translate what we
1196 * now know into VMA flags */
1197 vm_flags = determine_vm_flags(file, prot, flags, capabilities);
1198
David Howells8feae132009-01-08 12:04:47 +00001199 /* we're going to need to record the mapping */
1200 region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1201 if (!region)
1202 goto error_getting_region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203
David Howells8feae132009-01-08 12:04:47 +00001204 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1205 if (!vma)
1206 goto error_getting_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207
David Howells8feae132009-01-08 12:04:47 +00001208 atomic_set(&region->vm_usage, 1);
1209 region->vm_flags = vm_flags;
1210 region->vm_pgoff = pgoff;
1211
1212 INIT_LIST_HEAD(&vma->anon_vma_node);
1213 vma->vm_flags = vm_flags;
1214 vma->vm_pgoff = pgoff;
1215
1216 if (file) {
1217 region->vm_file = file;
1218 get_file(file);
1219 vma->vm_file = file;
1220 get_file(file);
1221 if (vm_flags & VM_EXECUTABLE) {
1222 added_exe_file_vma(current->mm);
1223 vma->vm_mm = current->mm;
1224 }
1225 }
1226
1227 down_write(&nommu_region_sem);
1228
1229 /* if we want to share, we need to check for regions created by other
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 * mmap() calls that overlap with our proposed mapping
David Howells8feae132009-01-08 12:04:47 +00001231 * - we can only share with a superset match on most regular files
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 * - shared mappings on character devices and memory backed files are
1233 * permitted to overlap inexactly as far as we are concerned for in
1234 * these cases, sharing is handled in the driver or filesystem rather
1235 * than here
1236 */
1237 if (vm_flags & VM_MAYSHARE) {
David Howells8feae132009-01-08 12:04:47 +00001238 struct vm_region *pregion;
1239 unsigned long pglen, rpglen, pgend, rpgend, start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240
David Howells8feae132009-01-08 12:04:47 +00001241 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1242 pgend = pgoff + pglen;
David Howells165b2392007-03-22 00:11:24 -08001243
David Howells8feae132009-01-08 12:04:47 +00001244 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1245 pregion = rb_entry(rb, struct vm_region, vm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246
David Howells8feae132009-01-08 12:04:47 +00001247 if (!(pregion->vm_flags & VM_MAYSHARE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248 continue;
1249
1250 /* search for overlapping mappings on the same file */
David Howells8feae132009-01-08 12:04:47 +00001251 if (pregion->vm_file->f_path.dentry->d_inode !=
1252 file->f_path.dentry->d_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 continue;
1254
David Howells8feae132009-01-08 12:04:47 +00001255 if (pregion->vm_pgoff >= pgend)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 continue;
1257
David Howells8feae132009-01-08 12:04:47 +00001258 rpglen = pregion->vm_end - pregion->vm_start;
1259 rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1260 rpgend = pregion->vm_pgoff + rpglen;
1261 if (pgoff >= rpgend)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 continue;
1263
David Howells8feae132009-01-08 12:04:47 +00001264 /* handle inexactly overlapping matches between
1265 * mappings */
1266 if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1267 !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1268 /* new mapping is not a subset of the region */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 if (!(capabilities & BDI_CAP_MAP_DIRECT))
1270 goto sharing_violation;
1271 continue;
1272 }
1273
David Howells8feae132009-01-08 12:04:47 +00001274 /* we've found a region we can share */
1275 atomic_inc(&pregion->vm_usage);
1276 vma->vm_region = pregion;
1277 start = pregion->vm_start;
1278 start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1279 vma->vm_start = start;
1280 vma->vm_end = start + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281
David Howells8feae132009-01-08 12:04:47 +00001282 if (pregion->vm_flags & VM_MAPPED_COPY) {
1283 kdebug("share copy");
1284 vma->vm_flags |= VM_MAPPED_COPY;
1285 } else {
1286 kdebug("share mmap");
1287 ret = do_mmap_shared_file(vma);
1288 if (ret < 0) {
1289 vma->vm_region = NULL;
1290 vma->vm_start = 0;
1291 vma->vm_end = 0;
1292 atomic_dec(&pregion->vm_usage);
1293 pregion = NULL;
1294 goto error_just_free;
1295 }
1296 }
1297 fput(region->vm_file);
1298 kmem_cache_free(vm_region_jar, region);
1299 region = pregion;
1300 result = start;
1301 goto share;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302 }
1303
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 /* obtain the address at which to make a shared mapping
1305 * - this is the hook for quasi-memory character devices to
1306 * tell us the location of a shared mapping
1307 */
1308 if (file && file->f_op->get_unmapped_area) {
1309 addr = file->f_op->get_unmapped_area(file, addr, len,
1310 pgoff, flags);
1311 if (IS_ERR((void *) addr)) {
1312 ret = addr;
1313 if (ret != (unsigned long) -ENOSYS)
David Howells8feae132009-01-08 12:04:47 +00001314 goto error_just_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315
1316 /* the driver refused to tell us where to site
1317 * the mapping so we'll have to attempt to copy
1318 * it */
1319 ret = (unsigned long) -ENODEV;
1320 if (!(capabilities & BDI_CAP_MAP_COPY))
David Howells8feae132009-01-08 12:04:47 +00001321 goto error_just_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322
1323 capabilities &= ~BDI_CAP_MAP_DIRECT;
David Howells8feae132009-01-08 12:04:47 +00001324 } else {
1325 vma->vm_start = region->vm_start = addr;
1326 vma->vm_end = region->vm_end = addr + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327 }
1328 }
1329 }
1330
David Howells8feae132009-01-08 12:04:47 +00001331 vma->vm_region = region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332
1333 /* set up the mapping */
1334 if (file && vma->vm_flags & VM_SHARED)
David Howells8feae132009-01-08 12:04:47 +00001335 ret = do_mmap_shared_file(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 else
David Howells8feae132009-01-08 12:04:47 +00001337 ret = do_mmap_private(vma, region, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 if (ret < 0)
David Howells8feae132009-01-08 12:04:47 +00001339 goto error_put_region;
1340
1341 add_nommu_region(region);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342
1343 /* okay... we have a mapping; now we have to register it */
David Howells8feae132009-01-08 12:04:47 +00001344 result = vma->vm_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 current->mm->total_vm += len >> PAGE_SHIFT;
1347
David Howells8feae132009-01-08 12:04:47 +00001348share:
1349 add_vma_to_mm(current->mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350
David Howells8feae132009-01-08 12:04:47 +00001351 up_write(&nommu_region_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352
1353 if (prot & PROT_EXEC)
David Howells8feae132009-01-08 12:04:47 +00001354 flush_icache_range(result, result + len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355
David Howells8feae132009-01-08 12:04:47 +00001356 kleave(" = %lx", result);
1357 return result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358
David Howells8feae132009-01-08 12:04:47 +00001359error_put_region:
1360 __put_nommu_region(region);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 if (vma) {
Matt Helsley925d1c42008-04-29 01:01:36 -07001362 if (vma->vm_file) {
Gavin Lambert3fcd03e2006-09-30 23:27:01 -07001363 fput(vma->vm_file);
Matt Helsley925d1c42008-04-29 01:01:36 -07001364 if (vma->vm_flags & VM_EXECUTABLE)
1365 removed_exe_file_vma(vma->vm_mm);
1366 }
David Howells8feae132009-01-08 12:04:47 +00001367 kmem_cache_free(vm_area_cachep, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368 }
David Howells8feae132009-01-08 12:04:47 +00001369 kleave(" = %d [pr]", ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 return ret;
1371
David Howells8feae132009-01-08 12:04:47 +00001372error_just_free:
1373 up_write(&nommu_region_sem);
1374error:
1375 fput(region->vm_file);
1376 kmem_cache_free(vm_region_jar, region);
1377 fput(vma->vm_file);
1378 if (vma->vm_flags & VM_EXECUTABLE)
1379 removed_exe_file_vma(vma->vm_mm);
1380 kmem_cache_free(vm_area_cachep, vma);
1381 kleave(" = %d", ret);
1382 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383
David Howells8feae132009-01-08 12:04:47 +00001384sharing_violation:
1385 up_write(&nommu_region_sem);
1386 printk(KERN_WARNING "Attempt to share mismatched mappings\n");
1387 ret = -EINVAL;
1388 goto error;
1389
1390error_getting_vma:
1391 kmem_cache_free(vm_region_jar, region);
1392 printk(KERN_WARNING "Allocation of vma for %lu byte allocation"
1393 " from process %d failed\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 len, current->pid);
1395 show_free_areas();
1396 return -ENOMEM;
1397
David Howells8feae132009-01-08 12:04:47 +00001398error_getting_region:
1399 printk(KERN_WARNING "Allocation of vm region for %lu byte allocation"
1400 " from process %d failed\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 len, current->pid);
1402 show_free_areas();
1403 return -ENOMEM;
1404}
Paul Mundtb5073172007-07-21 04:37:25 -07001405EXPORT_SYMBOL(do_mmap_pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406
1407/*
David Howells8feae132009-01-08 12:04:47 +00001408 * split a vma into two pieces at address 'addr', a new vma is allocated either
1409 * for the first part or the tail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 */
David Howells8feae132009-01-08 12:04:47 +00001411int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1412 unsigned long addr, int new_below)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413{
David Howells8feae132009-01-08 12:04:47 +00001414 struct vm_area_struct *new;
1415 struct vm_region *region;
1416 unsigned long npages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417
David Howells8feae132009-01-08 12:04:47 +00001418 kenter("");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419
David Howells8feae132009-01-08 12:04:47 +00001420 /* we're only permitted to split anonymous regions that have a single
1421 * owner */
1422 if (vma->vm_file ||
1423 atomic_read(&vma->vm_region->vm_usage) != 1)
1424 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425
David Howells8feae132009-01-08 12:04:47 +00001426 if (mm->map_count >= sysctl_max_map_count)
1427 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428
David Howells8feae132009-01-08 12:04:47 +00001429 region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1430 if (!region)
1431 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432
David Howells8feae132009-01-08 12:04:47 +00001433 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1434 if (!new) {
1435 kmem_cache_free(vm_region_jar, region);
1436 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437 }
David Howells8feae132009-01-08 12:04:47 +00001438
1439 /* most fields are the same, copy all, and then fixup */
1440 *new = *vma;
1441 *region = *vma->vm_region;
1442 new->vm_region = region;
1443
1444 npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1445
1446 if (new_below) {
Paul Mundtdd8632a2009-01-08 12:04:47 +00001447 region->vm_top = region->vm_end = new->vm_end = addr;
David Howells8feae132009-01-08 12:04:47 +00001448 } else {
1449 region->vm_start = new->vm_start = addr;
1450 region->vm_pgoff = new->vm_pgoff += npages;
1451 }
1452
1453 if (new->vm_ops && new->vm_ops->open)
1454 new->vm_ops->open(new);
1455
1456 delete_vma_from_mm(vma);
1457 down_write(&nommu_region_sem);
1458 delete_nommu_region(vma->vm_region);
1459 if (new_below) {
1460 vma->vm_region->vm_start = vma->vm_start = addr;
1461 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1462 } else {
1463 vma->vm_region->vm_end = vma->vm_end = addr;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001464 vma->vm_region->vm_top = addr;
David Howells8feae132009-01-08 12:04:47 +00001465 }
1466 add_nommu_region(vma->vm_region);
1467 add_nommu_region(new->vm_region);
1468 up_write(&nommu_region_sem);
1469 add_vma_to_mm(mm, vma);
1470 add_vma_to_mm(mm, new);
1471 return 0;
1472}
1473
1474/*
1475 * shrink a VMA by removing the specified chunk from either the beginning or
1476 * the end
1477 */
1478static int shrink_vma(struct mm_struct *mm,
1479 struct vm_area_struct *vma,
1480 unsigned long from, unsigned long to)
1481{
1482 struct vm_region *region;
1483
1484 kenter("");
1485
1486 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1487 * and list */
1488 delete_vma_from_mm(vma);
1489 if (from > vma->vm_start)
1490 vma->vm_end = from;
1491 else
1492 vma->vm_start = to;
1493 add_vma_to_mm(mm, vma);
1494
1495 /* cut the backing region down to size */
1496 region = vma->vm_region;
1497 BUG_ON(atomic_read(&region->vm_usage) != 1);
1498
1499 down_write(&nommu_region_sem);
1500 delete_nommu_region(region);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001501 if (from > region->vm_start) {
1502 to = region->vm_top;
1503 region->vm_top = region->vm_end = from;
1504 } else {
David Howells8feae132009-01-08 12:04:47 +00001505 region->vm_start = to;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001506 }
David Howells8feae132009-01-08 12:04:47 +00001507 add_nommu_region(region);
1508 up_write(&nommu_region_sem);
1509
1510 free_page_series(from, to);
1511 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512}
1513
David Howells30340972006-09-27 01:50:20 -07001514/*
1515 * release a mapping
David Howells8feae132009-01-08 12:04:47 +00001516 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1517 * VMA, though it need not cover the whole VMA
David Howells30340972006-09-27 01:50:20 -07001518 */
David Howells8feae132009-01-08 12:04:47 +00001519int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520{
David Howells8feae132009-01-08 12:04:47 +00001521 struct vm_area_struct *vma;
1522 struct rb_node *rb;
1523 unsigned long end = start + len;
1524 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525
David Howells8feae132009-01-08 12:04:47 +00001526 kenter(",%lx,%zx", start, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527
David Howells8feae132009-01-08 12:04:47 +00001528 if (len == 0)
1529 return -EINVAL;
1530
1531 /* find the first potentially overlapping VMA */
1532 vma = find_vma(mm, start);
1533 if (!vma) {
David Howells33e5d7692009-04-02 16:56:32 -07001534 static int limit = 0;
1535 if (limit < 5) {
1536 printk(KERN_WARNING
1537 "munmap of memory not mmapped by process %d"
1538 " (%s): 0x%lx-0x%lx\n",
1539 current->pid, current->comm,
1540 start, start + len - 1);
1541 limit++;
1542 }
David Howells8feae132009-01-08 12:04:47 +00001543 return -EINVAL;
David Howells30340972006-09-27 01:50:20 -07001544 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545
David Howells8feae132009-01-08 12:04:47 +00001546 /* we're allowed to split an anonymous VMA but not a file-backed one */
1547 if (vma->vm_file) {
1548 do {
1549 if (start > vma->vm_start) {
1550 kleave(" = -EINVAL [miss]");
1551 return -EINVAL;
1552 }
1553 if (end == vma->vm_end)
1554 goto erase_whole_vma;
1555 rb = rb_next(&vma->vm_rb);
1556 vma = rb_entry(rb, struct vm_area_struct, vm_rb);
1557 } while (rb);
1558 kleave(" = -EINVAL [split file]");
1559 return -EINVAL;
1560 } else {
1561 /* the chunk must be a subset of the VMA found */
1562 if (start == vma->vm_start && end == vma->vm_end)
1563 goto erase_whole_vma;
1564 if (start < vma->vm_start || end > vma->vm_end) {
1565 kleave(" = -EINVAL [superset]");
1566 return -EINVAL;
1567 }
1568 if (start & ~PAGE_MASK) {
1569 kleave(" = -EINVAL [unaligned start]");
1570 return -EINVAL;
1571 }
1572 if (end != vma->vm_end && end & ~PAGE_MASK) {
1573 kleave(" = -EINVAL [unaligned split]");
1574 return -EINVAL;
1575 }
1576 if (start != vma->vm_start && end != vma->vm_end) {
1577 ret = split_vma(mm, vma, start, 1);
1578 if (ret < 0) {
1579 kleave(" = %d [split]", ret);
1580 return ret;
1581 }
1582 }
1583 return shrink_vma(mm, vma, start, end);
1584 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585
David Howells8feae132009-01-08 12:04:47 +00001586erase_whole_vma:
1587 delete_vma_from_mm(vma);
1588 delete_vma(mm, vma);
1589 kleave(" = 0");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 return 0;
1591}
Paul Mundtb5073172007-07-21 04:37:25 -07001592EXPORT_SYMBOL(do_munmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593
Heiko Carstens6a6160a2009-01-14 14:14:15 +01001594SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
David Howells30340972006-09-27 01:50:20 -07001595{
1596 int ret;
1597 struct mm_struct *mm = current->mm;
1598
1599 down_write(&mm->mmap_sem);
1600 ret = do_munmap(mm, addr, len);
1601 up_write(&mm->mmap_sem);
1602 return ret;
1603}
1604
1605/*
David Howells8feae132009-01-08 12:04:47 +00001606 * release all the mappings made in a process's VM space
David Howells30340972006-09-27 01:50:20 -07001607 */
David Howells8feae132009-01-08 12:04:47 +00001608void exit_mmap(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609{
David Howells8feae132009-01-08 12:04:47 +00001610 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611
David Howells8feae132009-01-08 12:04:47 +00001612 if (!mm)
1613 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614
David Howells8feae132009-01-08 12:04:47 +00001615 kenter("");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616
David Howells8feae132009-01-08 12:04:47 +00001617 mm->total_vm = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618
David Howells8feae132009-01-08 12:04:47 +00001619 while ((vma = mm->mmap)) {
1620 mm->mmap = vma->vm_next;
1621 delete_vma_from_mm(vma);
1622 delete_vma(mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623 }
David Howells8feae132009-01-08 12:04:47 +00001624
1625 kleave("");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626}
1627
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628unsigned long do_brk(unsigned long addr, unsigned long len)
1629{
1630 return -ENOMEM;
1631}
1632
1633/*
David Howells6fa5f802006-09-27 01:50:21 -07001634 * expand (or shrink) an existing mapping, potentially moving it at the same
1635 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 *
David Howells6fa5f802006-09-27 01:50:21 -07001637 * under NOMMU conditions, we only permit changing a mapping's size, and only
David Howells8feae132009-01-08 12:04:47 +00001638 * as long as it stays within the region allocated by do_mmap_private() and the
1639 * block is not shareable
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 *
David Howells6fa5f802006-09-27 01:50:21 -07001641 * MREMAP_FIXED is not supported under NOMMU conditions
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642 */
1643unsigned long do_mremap(unsigned long addr,
1644 unsigned long old_len, unsigned long new_len,
1645 unsigned long flags, unsigned long new_addr)
1646{
David Howells6fa5f802006-09-27 01:50:21 -07001647 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648
1649 /* insanity checks first */
David Howells8feae132009-01-08 12:04:47 +00001650 if (old_len == 0 || new_len == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651 return (unsigned long) -EINVAL;
1652
David Howells8feae132009-01-08 12:04:47 +00001653 if (addr & ~PAGE_MASK)
1654 return -EINVAL;
1655
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656 if (flags & MREMAP_FIXED && new_addr != addr)
1657 return (unsigned long) -EINVAL;
1658
David Howells8feae132009-01-08 12:04:47 +00001659 vma = find_vma_exact(current->mm, addr, old_len);
David Howells6fa5f802006-09-27 01:50:21 -07001660 if (!vma)
1661 return (unsigned long) -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662
David Howells6fa5f802006-09-27 01:50:21 -07001663 if (vma->vm_end != vma->vm_start + old_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 return (unsigned long) -EFAULT;
1665
David Howells6fa5f802006-09-27 01:50:21 -07001666 if (vma->vm_flags & VM_MAYSHARE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 return (unsigned long) -EPERM;
1668
David Howells8feae132009-01-08 12:04:47 +00001669 if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 return (unsigned long) -ENOMEM;
1671
1672 /* all checks complete - do it */
David Howells6fa5f802006-09-27 01:50:21 -07001673 vma->vm_end = vma->vm_start + new_len;
David Howells6fa5f802006-09-27 01:50:21 -07001674 return vma->vm_start;
1675}
Paul Mundtb5073172007-07-21 04:37:25 -07001676EXPORT_SYMBOL(do_mremap);
David Howells6fa5f802006-09-27 01:50:21 -07001677
Heiko Carstens6a6160a2009-01-14 14:14:15 +01001678SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1679 unsigned long, new_len, unsigned long, flags,
1680 unsigned long, new_addr)
David Howells6fa5f802006-09-27 01:50:21 -07001681{
1682 unsigned long ret;
1683
1684 down_write(&current->mm->mmap_sem);
1685 ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1686 up_write(&current->mm->mmap_sem);
1687 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688}
1689
Linus Torvalds6aab3412005-11-28 14:34:23 -08001690struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001691 unsigned int foll_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692{
1693 return NULL;
1694}
1695
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696int remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
1697 unsigned long to, unsigned long size, pgprot_t prot)
1698{
Greg Ungerer66aa2b42005-09-12 11:18:10 +10001699 vma->vm_start = vma->vm_pgoff << PAGE_SHIFT;
1700 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701}
Luke Yang22c4af42006-07-14 00:24:09 -07001702EXPORT_SYMBOL(remap_pfn_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703
Paul Mundtf905bc42008-02-04 22:29:59 -08001704int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1705 unsigned long pgoff)
1706{
1707 unsigned int size = vma->vm_end - vma->vm_start;
1708
1709 if (!(vma->vm_flags & VM_USERMAP))
1710 return -EINVAL;
1711
1712 vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1713 vma->vm_end = vma->vm_start + size;
1714
1715 return 0;
1716}
1717EXPORT_SYMBOL(remap_vmalloc_range);
1718
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1720{
1721}
1722
1723unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1724 unsigned long len, unsigned long pgoff, unsigned long flags)
1725{
1726 return -ENOMEM;
1727}
1728
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001729void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730{
1731}
1732
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733void unmap_mapping_range(struct address_space *mapping,
1734 loff_t const holebegin, loff_t const holelen,
1735 int even_cows)
1736{
1737}
Luke Yang22c4af42006-07-14 00:24:09 -07001738EXPORT_SYMBOL(unmap_mapping_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739
1740/*
David Howellsd56e03c2007-03-22 00:11:23 -08001741 * ask for an unmapped area at which to create a mapping on a file
1742 */
1743unsigned long get_unmapped_area(struct file *file, unsigned long addr,
1744 unsigned long len, unsigned long pgoff,
1745 unsigned long flags)
1746{
1747 unsigned long (*get_area)(struct file *, unsigned long, unsigned long,
1748 unsigned long, unsigned long);
1749
1750 get_area = current->mm->get_unmapped_area;
1751 if (file && file->f_op && file->f_op->get_unmapped_area)
1752 get_area = file->f_op->get_unmapped_area;
1753
1754 if (!get_area)
1755 return -ENOSYS;
1756
1757 return get_area(file, addr, len, pgoff, flags);
1758}
David Howellsd56e03c2007-03-22 00:11:23 -08001759EXPORT_SYMBOL(get_unmapped_area);
1760
1761/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 * Check that a process has enough memory to allocate a new virtual
1763 * mapping. 0 means there is enough memory for the allocation to
1764 * succeed and -ENOMEM implies there is not.
1765 *
1766 * We currently support three overcommit policies, which are set via the
1767 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
1768 *
1769 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1770 * Additional code 2002 Jul 20 by Robert Love.
1771 *
1772 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1773 *
1774 * Note this is a helper function intended to be used by LSMs which
1775 * wish to use this logic.
1776 */
Alan Cox34b4e4a2007-08-22 14:01:28 -07001777int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778{
1779 unsigned long free, allowed;
1780
1781 vm_acct_memory(pages);
1782
1783 /*
1784 * Sometimes we want to use more memory than we have
1785 */
1786 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1787 return 0;
1788
1789 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1790 unsigned long n;
1791
Christoph Lameter347ce432006-06-30 01:55:35 -07001792 free = global_page_state(NR_FILE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793 free += nr_swap_pages;
1794
1795 /*
1796 * Any slabs which are created with the
1797 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1798 * which are reclaimable, under pressure. The dentry
1799 * cache and most inode caches should fall into this
1800 */
Christoph Lameter972d1a72006-09-25 23:31:51 -07001801 free += global_page_state(NR_SLAB_RECLAIMABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802
1803 /*
1804 * Leave the last 3% for root
1805 */
1806 if (!cap_sys_admin)
1807 free -= free / 32;
1808
1809 if (free > pages)
1810 return 0;
1811
1812 /*
1813 * nr_free_pages() is very expensive on large systems,
1814 * only call if we're about to fail.
1815 */
1816 n = nr_free_pages();
Hideo AOKId5ddc792006-04-10 22:53:01 -07001817
1818 /*
1819 * Leave reserved pages. The pages are not for anonymous pages.
1820 */
1821 if (n <= totalreserve_pages)
1822 goto error;
1823 else
1824 n -= totalreserve_pages;
1825
1826 /*
1827 * Leave the last 3% for root
1828 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829 if (!cap_sys_admin)
1830 n -= n / 32;
1831 free += n;
1832
1833 if (free > pages)
1834 return 0;
Hideo AOKId5ddc792006-04-10 22:53:01 -07001835
1836 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837 }
1838
1839 allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1840 /*
1841 * Leave the last 3% for root
1842 */
1843 if (!cap_sys_admin)
1844 allowed -= allowed / 32;
1845 allowed += total_swap_pages;
1846
1847 /* Don't let a single process grow too big:
1848 leave 3% of the size of this process for other processes */
Alan Cox731572d2008-10-29 14:01:20 -07001849 if (mm)
1850 allowed -= mm->total_vm / 32;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07001852 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853 return 0;
KOSAKI Motohiro00a62ce2009-04-30 15:08:51 -07001854
Hideo AOKId5ddc792006-04-10 22:53:01 -07001855error:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856 vm_unacct_memory(pages);
1857
1858 return -ENOMEM;
1859}
1860
1861int in_gate_area_no_task(unsigned long addr)
1862{
1863 return 0;
1864}
David Howellsb0e15192006-01-06 00:11:42 -08001865
Nick Piggind0217ac2007-07-19 01:47:03 -07001866int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
David Howellsb0e15192006-01-06 00:11:42 -08001867{
1868 BUG();
Nick Piggind0217ac2007-07-19 01:47:03 -07001869 return 0;
David Howellsb0e15192006-01-06 00:11:42 -08001870}
Paul Mundtb5073172007-07-21 04:37:25 -07001871EXPORT_SYMBOL(filemap_fault);
David Howells0ec76a12006-09-27 01:50:15 -07001872
1873/*
1874 * Access another process' address space.
1875 * - source/target buffer must be kernel space
1876 */
1877int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
1878{
David Howells0ec76a12006-09-27 01:50:15 -07001879 struct vm_area_struct *vma;
1880 struct mm_struct *mm;
1881
1882 if (addr + len < addr)
1883 return 0;
1884
1885 mm = get_task_mm(tsk);
1886 if (!mm)
1887 return 0;
1888
1889 down_read(&mm->mmap_sem);
1890
1891 /* the access must start within one of the target process's mappings */
David Howells0159b142006-09-27 01:50:16 -07001892 vma = find_vma(mm, addr);
1893 if (vma) {
David Howells0ec76a12006-09-27 01:50:15 -07001894 /* don't overrun this mapping */
1895 if (addr + len >= vma->vm_end)
1896 len = vma->vm_end - addr;
1897
1898 /* only read or write mappings where it is permitted */
David Howellsd00c7b92006-09-27 01:50:19 -07001899 if (write && vma->vm_flags & VM_MAYWRITE)
David Howells0ec76a12006-09-27 01:50:15 -07001900 len -= copy_to_user((void *) addr, buf, len);
David Howellsd00c7b92006-09-27 01:50:19 -07001901 else if (!write && vma->vm_flags & VM_MAYREAD)
David Howells0ec76a12006-09-27 01:50:15 -07001902 len -= copy_from_user(buf, (void *) addr, len);
1903 else
1904 len = 0;
1905 } else {
1906 len = 0;
1907 }
1908
1909 up_read(&mm->mmap_sem);
1910 mmput(mm);
1911 return len;
1912}