blob: 2fcf47d449b4e203d46bbbf76827aeedbd2e2c22 [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;
Alan Cox80119ef2008-05-23 13:04:31 -070065atomic_long_t vm_committed_space = ATOMIC_LONG_INIT(0);
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 Howells8feae132009-01-08 12:04:47 +000072atomic_t mmap_pages_allocated;
73
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{
David Howells8feae132009-01-08 12:04:47 +0000466 vm_region_jar = kmem_cache_create("vm_region_jar",
467 sizeof(struct vm_region), 0,
468 SLAB_PANIC, NULL);
469 vm_area_cachep = kmem_cache_create("vm_area_struct",
470 sizeof(struct vm_area_struct), 0,
471 SLAB_PANIC, NULL);
472}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473
David Howells8feae132009-01-08 12:04:47 +0000474/*
475 * validate the region tree
476 * - the caller must hold the region lock
477 */
478#ifdef CONFIG_DEBUG_NOMMU_REGIONS
479static noinline void validate_nommu_regions(void)
480{
481 struct vm_region *region, *last;
482 struct rb_node *p, *lastp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483
David Howells8feae132009-01-08 12:04:47 +0000484 lastp = rb_first(&nommu_region_tree);
485 if (!lastp)
486 return;
487
488 last = rb_entry(lastp, struct vm_region, vm_rb);
489 if (unlikely(last->vm_end <= last->vm_start))
490 BUG();
Paul Mundtdd8632a2009-01-08 12:04:47 +0000491 if (unlikely(last->vm_top < last->vm_end))
492 BUG();
David Howells8feae132009-01-08 12:04:47 +0000493
494 while ((p = rb_next(lastp))) {
495 region = rb_entry(p, struct vm_region, vm_rb);
496 last = rb_entry(lastp, struct vm_region, vm_rb);
497
498 if (unlikely(region->vm_end <= region->vm_start))
499 BUG();
Paul Mundtdd8632a2009-01-08 12:04:47 +0000500 if (unlikely(region->vm_top < region->vm_end))
501 BUG();
502 if (unlikely(region->vm_start < last->vm_top))
David Howells8feae132009-01-08 12:04:47 +0000503 BUG();
504
505 lastp = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 }
507}
David Howells8feae132009-01-08 12:04:47 +0000508#else
509#define validate_nommu_regions() do {} while(0)
510#endif
511
512/*
513 * add a region into the global tree
514 */
515static void add_nommu_region(struct vm_region *region)
516{
517 struct vm_region *pregion;
518 struct rb_node **p, *parent;
519
520 validate_nommu_regions();
521
522 BUG_ON(region->vm_start & ~PAGE_MASK);
523
524 parent = NULL;
525 p = &nommu_region_tree.rb_node;
526 while (*p) {
527 parent = *p;
528 pregion = rb_entry(parent, struct vm_region, vm_rb);
529 if (region->vm_start < pregion->vm_start)
530 p = &(*p)->rb_left;
531 else if (region->vm_start > pregion->vm_start)
532 p = &(*p)->rb_right;
533 else if (pregion == region)
534 return;
535 else
536 BUG();
537 }
538
539 rb_link_node(&region->vm_rb, parent, p);
540 rb_insert_color(&region->vm_rb, &nommu_region_tree);
541
542 validate_nommu_regions();
543}
544
545/*
546 * delete a region from the global tree
547 */
548static void delete_nommu_region(struct vm_region *region)
549{
550 BUG_ON(!nommu_region_tree.rb_node);
551
552 validate_nommu_regions();
553 rb_erase(&region->vm_rb, &nommu_region_tree);
554 validate_nommu_regions();
555}
556
557/*
558 * free a contiguous series of pages
559 */
560static void free_page_series(unsigned long from, unsigned long to)
561{
562 for (; from < to; from += PAGE_SIZE) {
563 struct page *page = virt_to_page(from);
564
565 kdebug("- free %lx", from);
566 atomic_dec(&mmap_pages_allocated);
567 if (page_count(page) != 1)
568 kdebug("free page %p [%d]", page, page_count(page));
569 put_page(page);
570 }
571}
572
573/*
574 * release a reference to a region
575 * - the caller must hold the region semaphore, which this releases
Paul Mundtdd8632a2009-01-08 12:04:47 +0000576 * - the region may not have been added to the tree yet, in which case vm_top
David Howells8feae132009-01-08 12:04:47 +0000577 * will equal vm_start
578 */
579static void __put_nommu_region(struct vm_region *region)
580 __releases(nommu_region_sem)
581{
582 kenter("%p{%d}", region, atomic_read(&region->vm_usage));
583
584 BUG_ON(!nommu_region_tree.rb_node);
585
586 if (atomic_dec_and_test(&region->vm_usage)) {
Paul Mundtdd8632a2009-01-08 12:04:47 +0000587 if (region->vm_top > region->vm_start)
David Howells8feae132009-01-08 12:04:47 +0000588 delete_nommu_region(region);
589 up_write(&nommu_region_sem);
590
591 if (region->vm_file)
592 fput(region->vm_file);
593
594 /* IO memory and memory shared directly out of the pagecache
595 * from ramfs/tmpfs mustn't be released here */
596 if (region->vm_flags & VM_MAPPED_COPY) {
597 kdebug("free series");
Paul Mundtdd8632a2009-01-08 12:04:47 +0000598 free_page_series(region->vm_start, region->vm_top);
David Howells8feae132009-01-08 12:04:47 +0000599 }
600 kmem_cache_free(vm_region_jar, region);
601 } else {
602 up_write(&nommu_region_sem);
603 }
604}
605
606/*
607 * release a reference to a region
608 */
609static void put_nommu_region(struct vm_region *region)
610{
611 down_write(&nommu_region_sem);
612 __put_nommu_region(region);
613}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614
David Howells30340972006-09-27 01:50:20 -0700615/*
616 * add a VMA into a process's mm_struct in the appropriate place in the list
David Howells8feae132009-01-08 12:04:47 +0000617 * and tree and add to the address space's page tree also if not an anonymous
618 * page
David Howells30340972006-09-27 01:50:20 -0700619 * - should be called with mm->mmap_sem held writelocked
620 */
David Howells8feae132009-01-08 12:04:47 +0000621static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
David Howells30340972006-09-27 01:50:20 -0700622{
David Howells8feae132009-01-08 12:04:47 +0000623 struct vm_area_struct *pvma, **pp;
624 struct address_space *mapping;
625 struct rb_node **p, *parent;
David Howells30340972006-09-27 01:50:20 -0700626
David Howells8feae132009-01-08 12:04:47 +0000627 kenter(",%p", vma);
628
629 BUG_ON(!vma->vm_region);
630
631 mm->map_count++;
632 vma->vm_mm = mm;
633
634 /* add the VMA to the mapping */
635 if (vma->vm_file) {
636 mapping = vma->vm_file->f_mapping;
637
638 flush_dcache_mmap_lock(mapping);
639 vma_prio_tree_insert(vma, &mapping->i_mmap);
640 flush_dcache_mmap_unlock(mapping);
641 }
642
643 /* add the VMA to the tree */
644 parent = NULL;
645 p = &mm->mm_rb.rb_node;
646 while (*p) {
647 parent = *p;
648 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
649
650 /* sort by: start addr, end addr, VMA struct addr in that order
651 * (the latter is necessary as we may get identical VMAs) */
652 if (vma->vm_start < pvma->vm_start)
653 p = &(*p)->rb_left;
654 else if (vma->vm_start > pvma->vm_start)
655 p = &(*p)->rb_right;
656 else if (vma->vm_end < pvma->vm_end)
657 p = &(*p)->rb_left;
658 else if (vma->vm_end > pvma->vm_end)
659 p = &(*p)->rb_right;
660 else if (vma < pvma)
661 p = &(*p)->rb_left;
662 else if (vma > pvma)
663 p = &(*p)->rb_right;
664 else
665 BUG();
666 }
667
668 rb_link_node(&vma->vm_rb, parent, p);
669 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
670
671 /* add VMA to the VMA list also */
672 for (pp = &mm->mmap; (pvma = *pp); pp = &(*pp)->vm_next) {
673 if (pvma->vm_start > vma->vm_start)
David Howells30340972006-09-27 01:50:20 -0700674 break;
David Howells8feae132009-01-08 12:04:47 +0000675 if (pvma->vm_start < vma->vm_start)
676 continue;
677 if (pvma->vm_end < vma->vm_end)
678 break;
679 }
David Howells30340972006-09-27 01:50:20 -0700680
David Howells8feae132009-01-08 12:04:47 +0000681 vma->vm_next = *pp;
682 *pp = vma;
683}
684
685/*
686 * delete a VMA from its owning mm_struct and address space
687 */
688static void delete_vma_from_mm(struct vm_area_struct *vma)
689{
690 struct vm_area_struct **pp;
691 struct address_space *mapping;
692 struct mm_struct *mm = vma->vm_mm;
693
694 kenter("%p", vma);
695
696 mm->map_count--;
697 if (mm->mmap_cache == vma)
698 mm->mmap_cache = NULL;
699
700 /* remove the VMA from the mapping */
701 if (vma->vm_file) {
702 mapping = vma->vm_file->f_mapping;
703
704 flush_dcache_mmap_lock(mapping);
705 vma_prio_tree_remove(vma, &mapping->i_mmap);
706 flush_dcache_mmap_unlock(mapping);
707 }
708
709 /* remove from the MM's tree and list */
710 rb_erase(&vma->vm_rb, &mm->mm_rb);
711 for (pp = &mm->mmap; *pp; pp = &(*pp)->vm_next) {
712 if (*pp == vma) {
713 *pp = vma->vm_next;
714 break;
715 }
716 }
717
718 vma->vm_mm = NULL;
719}
720
721/*
722 * destroy a VMA record
723 */
724static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
725{
726 kenter("%p", vma);
727 if (vma->vm_ops && vma->vm_ops->close)
728 vma->vm_ops->close(vma);
729 if (vma->vm_file) {
730 fput(vma->vm_file);
731 if (vma->vm_flags & VM_EXECUTABLE)
732 removed_exe_file_vma(mm);
733 }
734 put_nommu_region(vma->vm_region);
735 kmem_cache_free(vm_area_cachep, vma);
David Howells30340972006-09-27 01:50:20 -0700736}
737
738/*
739 * look up the first VMA in which addr resides, NULL if none
740 * - should be called with mm->mmap_sem at least held readlocked
741 */
742struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
743{
David Howells8feae132009-01-08 12:04:47 +0000744 struct vm_area_struct *vma;
745 struct rb_node *n = mm->mm_rb.rb_node;
David Howells30340972006-09-27 01:50:20 -0700746
David Howells8feae132009-01-08 12:04:47 +0000747 /* check the cache first */
748 vma = mm->mmap_cache;
749 if (vma && vma->vm_start <= addr && vma->vm_end > addr)
750 return vma;
751
752 /* trawl the tree (there may be multiple mappings in which addr
753 * resides) */
754 for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) {
755 vma = rb_entry(n, struct vm_area_struct, vm_rb);
756 if (vma->vm_start > addr)
757 return NULL;
758 if (vma->vm_end > addr) {
759 mm->mmap_cache = vma;
760 return vma;
761 }
David Howells30340972006-09-27 01:50:20 -0700762 }
763
David Howells30340972006-09-27 01:50:20 -0700764 return NULL;
765}
766EXPORT_SYMBOL(find_vma);
767
768/*
David Howells930e6522006-09-27 01:50:22 -0700769 * find a VMA
770 * - we don't extend stack VMAs under NOMMU conditions
771 */
772struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
773{
774 return find_vma(mm, addr);
775}
776
David Howells8feae132009-01-08 12:04:47 +0000777/*
778 * expand a stack to a given address
779 * - not supported under NOMMU conditions
780 */
Greg Ungerer57c8f632007-07-15 23:38:28 -0700781int expand_stack(struct vm_area_struct *vma, unsigned long address)
782{
783 return -ENOMEM;
784}
785
David Howells930e6522006-09-27 01:50:22 -0700786/*
David Howells6fa5f802006-09-27 01:50:21 -0700787 * look up the first VMA exactly that exactly matches addr
788 * - should be called with mm->mmap_sem at least held readlocked
789 */
David Howells8feae132009-01-08 12:04:47 +0000790static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
791 unsigned long addr,
792 unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793{
794 struct vm_area_struct *vma;
David Howells8feae132009-01-08 12:04:47 +0000795 struct rb_node *n = mm->mm_rb.rb_node;
796 unsigned long end = addr + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797
David Howells8feae132009-01-08 12:04:47 +0000798 /* check the cache first */
799 vma = mm->mmap_cache;
800 if (vma && vma->vm_start == addr && vma->vm_end == end)
801 return vma;
802
803 /* trawl the tree (there may be multiple mappings in which addr
804 * resides) */
805 for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 vma = rb_entry(n, struct vm_area_struct, vm_rb);
David Howells8feae132009-01-08 12:04:47 +0000807 if (vma->vm_start < addr)
808 continue;
809 if (vma->vm_start > addr)
810 return NULL;
811 if (vma->vm_end == end) {
812 mm->mmap_cache = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 return vma;
David Howells8feae132009-01-08 12:04:47 +0000814 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 }
816
817 return NULL;
818}
819
David Howells30340972006-09-27 01:50:20 -0700820/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 * determine whether a mapping should be permitted and, if so, what sort of
822 * mapping we're capable of supporting
823 */
824static int validate_mmap_request(struct file *file,
825 unsigned long addr,
826 unsigned long len,
827 unsigned long prot,
828 unsigned long flags,
829 unsigned long pgoff,
830 unsigned long *_capabilities)
831{
David Howells8feae132009-01-08 12:04:47 +0000832 unsigned long capabilities, rlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 unsigned long reqprot = prot;
834 int ret;
835
836 /* do the simple checks first */
837 if (flags & MAP_FIXED || addr) {
838 printk(KERN_DEBUG
839 "%d: Can't do fixed-address/overlay mmap of RAM\n",
840 current->pid);
841 return -EINVAL;
842 }
843
844 if ((flags & MAP_TYPE) != MAP_PRIVATE &&
845 (flags & MAP_TYPE) != MAP_SHARED)
846 return -EINVAL;
847
Mike Frysingerf81cff02006-12-06 12:02:59 +1000848 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 return -EINVAL;
850
Mike Frysingerf81cff02006-12-06 12:02:59 +1000851 /* Careful about overflows.. */
David Howells8feae132009-01-08 12:04:47 +0000852 rlen = PAGE_ALIGN(len);
853 if (!rlen || rlen > TASK_SIZE)
Mike Frysingerf81cff02006-12-06 12:02:59 +1000854 return -ENOMEM;
855
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 /* offset overflow? */
David Howells8feae132009-01-08 12:04:47 +0000857 if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
Mike Frysingerf81cff02006-12-06 12:02:59 +1000858 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859
860 if (file) {
861 /* validate file mapping requests */
862 struct address_space *mapping;
863
864 /* files must support mmap */
865 if (!file->f_op || !file->f_op->mmap)
866 return -ENODEV;
867
868 /* work out if what we've got could possibly be shared
869 * - we support chardevs that provide their own "memory"
870 * - we support files/blockdevs that are memory backed
871 */
872 mapping = file->f_mapping;
873 if (!mapping)
Josef Sipeke9536ae2006-12-08 02:37:21 -0800874 mapping = file->f_path.dentry->d_inode->i_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875
876 capabilities = 0;
877 if (mapping && mapping->backing_dev_info)
878 capabilities = mapping->backing_dev_info->capabilities;
879
880 if (!capabilities) {
881 /* no explicit capabilities set, so assume some
882 * defaults */
Josef Sipeke9536ae2006-12-08 02:37:21 -0800883 switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 case S_IFREG:
885 case S_IFBLK:
886 capabilities = BDI_CAP_MAP_COPY;
887 break;
888
889 case S_IFCHR:
890 capabilities =
891 BDI_CAP_MAP_DIRECT |
892 BDI_CAP_READ_MAP |
893 BDI_CAP_WRITE_MAP;
894 break;
895
896 default:
897 return -EINVAL;
898 }
899 }
900
901 /* eliminate any capabilities that we can't support on this
902 * device */
903 if (!file->f_op->get_unmapped_area)
904 capabilities &= ~BDI_CAP_MAP_DIRECT;
905 if (!file->f_op->read)
906 capabilities &= ~BDI_CAP_MAP_COPY;
907
908 if (flags & MAP_SHARED) {
909 /* do checks for writing, appending and locking */
910 if ((prot & PROT_WRITE) &&
911 !(file->f_mode & FMODE_WRITE))
912 return -EACCES;
913
Josef Sipeke9536ae2006-12-08 02:37:21 -0800914 if (IS_APPEND(file->f_path.dentry->d_inode) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 (file->f_mode & FMODE_WRITE))
916 return -EACCES;
917
Josef Sipeke9536ae2006-12-08 02:37:21 -0800918 if (locks_verify_locked(file->f_path.dentry->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 return -EAGAIN;
920
921 if (!(capabilities & BDI_CAP_MAP_DIRECT))
922 return -ENODEV;
923
924 if (((prot & PROT_READ) && !(capabilities & BDI_CAP_READ_MAP)) ||
925 ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
926 ((prot & PROT_EXEC) && !(capabilities & BDI_CAP_EXEC_MAP))
927 ) {
928 printk("MAP_SHARED not completely supported on !MMU\n");
929 return -EINVAL;
930 }
931
932 /* we mustn't privatise shared mappings */
933 capabilities &= ~BDI_CAP_MAP_COPY;
934 }
935 else {
936 /* we're going to read the file into private memory we
937 * allocate */
938 if (!(capabilities & BDI_CAP_MAP_COPY))
939 return -ENODEV;
940
941 /* we don't permit a private writable mapping to be
942 * shared with the backing device */
943 if (prot & PROT_WRITE)
944 capabilities &= ~BDI_CAP_MAP_DIRECT;
945 }
946
947 /* handle executable mappings and implied executable
948 * mappings */
Josef Sipeke9536ae2006-12-08 02:37:21 -0800949 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950 if (prot & PROT_EXEC)
951 return -EPERM;
952 }
953 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
954 /* handle implication of PROT_EXEC by PROT_READ */
955 if (current->personality & READ_IMPLIES_EXEC) {
956 if (capabilities & BDI_CAP_EXEC_MAP)
957 prot |= PROT_EXEC;
958 }
959 }
960 else if ((prot & PROT_READ) &&
961 (prot & PROT_EXEC) &&
962 !(capabilities & BDI_CAP_EXEC_MAP)
963 ) {
964 /* backing file is not executable, try to copy */
965 capabilities &= ~BDI_CAP_MAP_DIRECT;
966 }
967 }
968 else {
969 /* anonymous mappings are always memory backed and can be
970 * privately mapped
971 */
972 capabilities = BDI_CAP_MAP_COPY;
973
974 /* handle PROT_EXEC implication by PROT_READ */
975 if ((prot & PROT_READ) &&
976 (current->personality & READ_IMPLIES_EXEC))
977 prot |= PROT_EXEC;
978 }
979
980 /* allow the security API to have its say */
Eric Parised032182007-06-28 15:55:21 -0400981 ret = security_file_mmap(file, reqprot, prot, flags, addr, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 if (ret < 0)
983 return ret;
984
985 /* looks okay */
986 *_capabilities = capabilities;
987 return 0;
988}
989
990/*
991 * we've determined that we can make the mapping, now translate what we
992 * now know into VMA flags
993 */
994static unsigned long determine_vm_flags(struct file *file,
995 unsigned long prot,
996 unsigned long flags,
997 unsigned long capabilities)
998{
999 unsigned long vm_flags;
1000
1001 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
1002 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1003 /* vm_flags |= mm->def_flags; */
1004
1005 if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
1006 /* attempt to share read-only copies of mapped file chunks */
1007 if (file && !(prot & PROT_WRITE))
1008 vm_flags |= VM_MAYSHARE;
1009 }
1010 else {
1011 /* overlay a shareable mapping on the backing device or inode
1012 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1013 * romfs/cramfs */
1014 if (flags & MAP_SHARED)
1015 vm_flags |= VM_MAYSHARE | VM_SHARED;
1016 else if ((((vm_flags & capabilities) ^ vm_flags) & BDI_CAP_VMFLAGS) == 0)
1017 vm_flags |= VM_MAYSHARE;
1018 }
1019
1020 /* refuse to let anyone share private mappings with this process if
1021 * it's being traced - otherwise breakpoints set in it may interfere
1022 * with another untraced process
1023 */
Roland McGrathfa8e26c2008-07-25 19:45:50 -07001024 if ((flags & MAP_PRIVATE) && tracehook_expect_breakpoints(current))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 vm_flags &= ~VM_MAYSHARE;
1026
1027 return vm_flags;
1028}
1029
1030/*
David Howells8feae132009-01-08 12:04:47 +00001031 * set up a shared mapping on a file (the driver or filesystem provides and
1032 * pins the storage)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 */
David Howells8feae132009-01-08 12:04:47 +00001034static int do_mmap_shared_file(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035{
1036 int ret;
1037
1038 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001039 if (ret == 0) {
1040 vma->vm_region->vm_top = vma->vm_region->vm_end;
1041 return ret;
1042 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 if (ret != -ENOSYS)
1044 return ret;
1045
1046 /* getting an ENOSYS error indicates that direct mmap isn't
1047 * possible (as opposed to tried but failed) so we'll fall
1048 * through to making a private copy of the data and mapping
1049 * that if we can */
1050 return -ENODEV;
1051}
1052
1053/*
1054 * set up a private mapping or an anonymous shared mapping
1055 */
David Howells8feae132009-01-08 12:04:47 +00001056static int do_mmap_private(struct vm_area_struct *vma,
1057 struct vm_region *region,
1058 unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059{
David Howells8feae132009-01-08 12:04:47 +00001060 struct page *pages;
1061 unsigned long total, point, n, rlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062 void *base;
David Howells8feae132009-01-08 12:04:47 +00001063 int ret, order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064
1065 /* invoke the file's mapping function so that it can keep track of
1066 * shared mappings on devices or memory
1067 * - VM_MAYSHARE will be set if it may attempt to share
1068 */
1069 if (vma->vm_file) {
1070 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001071 if (ret == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 /* shouldn't return success if we're not sharing */
Paul Mundtdd8632a2009-01-08 12:04:47 +00001073 BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1074 vma->vm_region->vm_top = vma->vm_region->vm_end;
1075 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 }
Paul Mundtdd8632a2009-01-08 12:04:47 +00001077 if (ret != -ENOSYS)
1078 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079
1080 /* getting an ENOSYS error indicates that direct mmap isn't
1081 * possible (as opposed to tried but failed) so we'll try to
1082 * make a private copy of the data and map that instead */
1083 }
1084
David Howells8feae132009-01-08 12:04:47 +00001085 rlen = PAGE_ALIGN(len);
1086
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 /* allocate some memory to hold the mapping
1088 * - note that this may not return a page-aligned address if the object
1089 * we're allocating is smaller than a page
1090 */
David Howells8feae132009-01-08 12:04:47 +00001091 order = get_order(rlen);
1092 kdebug("alloc order %d for %lx", order, len);
1093
1094 pages = alloc_pages(GFP_KERNEL, order);
1095 if (!pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 goto enomem;
1097
David Howells8feae132009-01-08 12:04:47 +00001098 total = 1 << order;
1099 atomic_add(total, &mmap_pages_allocated);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100
David Howells8feae132009-01-08 12:04:47 +00001101 point = rlen >> PAGE_SHIFT;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001102
1103 /* we allocated a power-of-2 sized page set, so we may want to trim off
1104 * the excess */
1105 if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) {
1106 while (total > point) {
1107 order = ilog2(total - point);
1108 n = 1 << order;
1109 kdebug("shave %lu/%lu @%lu", n, total - point, total);
1110 atomic_sub(n, &mmap_pages_allocated);
1111 total -= n;
1112 set_page_refcounted(pages + total);
1113 __free_pages(pages + total, order);
1114 }
David Howells8feae132009-01-08 12:04:47 +00001115 }
1116
David Howells8feae132009-01-08 12:04:47 +00001117 for (point = 1; point < total; point++)
1118 set_page_refcounted(&pages[point]);
1119
1120 base = page_address(pages);
1121 region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1122 region->vm_start = (unsigned long) base;
1123 region->vm_end = region->vm_start + rlen;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001124 region->vm_top = region->vm_start + (total << PAGE_SHIFT);
David Howells8feae132009-01-08 12:04:47 +00001125
1126 vma->vm_start = region->vm_start;
1127 vma->vm_end = region->vm_start + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128
1129 if (vma->vm_file) {
1130 /* read the contents of a file into the copy */
1131 mm_segment_t old_fs;
1132 loff_t fpos;
1133
1134 fpos = vma->vm_pgoff;
1135 fpos <<= PAGE_SHIFT;
1136
1137 old_fs = get_fs();
1138 set_fs(KERNEL_DS);
David Howells8feae132009-01-08 12:04:47 +00001139 ret = vma->vm_file->f_op->read(vma->vm_file, base, rlen, &fpos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 set_fs(old_fs);
1141
1142 if (ret < 0)
1143 goto error_free;
1144
1145 /* clear the last little bit */
David Howells8feae132009-01-08 12:04:47 +00001146 if (ret < rlen)
1147 memset(base + ret, 0, rlen - ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148
1149 } else {
1150 /* if it's an anonymous mapping, then just clear it */
David Howells8feae132009-01-08 12:04:47 +00001151 memset(base, 0, rlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 }
1153
1154 return 0;
1155
1156error_free:
David Howells8feae132009-01-08 12:04:47 +00001157 free_page_series(region->vm_start, region->vm_end);
1158 region->vm_start = vma->vm_start = 0;
1159 region->vm_end = vma->vm_end = 0;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001160 region->vm_top = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 return ret;
1162
1163enomem:
Greg Ungerer05ae6fa2009-01-13 17:30:22 +10001164 printk("Allocation of length %lu from process %d (%s) failed\n",
1165 len, current->pid, current->comm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 show_free_areas();
1167 return -ENOMEM;
1168}
1169
1170/*
1171 * handle mapping creation for uClinux
1172 */
1173unsigned long do_mmap_pgoff(struct file *file,
1174 unsigned long addr,
1175 unsigned long len,
1176 unsigned long prot,
1177 unsigned long flags,
1178 unsigned long pgoff)
1179{
David Howells8feae132009-01-08 12:04:47 +00001180 struct vm_area_struct *vma;
1181 struct vm_region *region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 struct rb_node *rb;
David Howells8feae132009-01-08 12:04:47 +00001183 unsigned long capabilities, vm_flags, result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 int ret;
1185
David Howells8feae132009-01-08 12:04:47 +00001186 kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff);
1187
Eric Paris7cd94142007-11-26 18:47:40 -05001188 if (!(flags & MAP_FIXED))
1189 addr = round_hint_to_min(addr);
1190
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191 /* decide whether we should attempt the mapping, and if so what sort of
1192 * mapping */
1193 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1194 &capabilities);
David Howells8feae132009-01-08 12:04:47 +00001195 if (ret < 0) {
1196 kleave(" = %d [val]", ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 return ret;
David Howells8feae132009-01-08 12:04:47 +00001198 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199
1200 /* we've determined that we can make the mapping, now translate what we
1201 * now know into VMA flags */
1202 vm_flags = determine_vm_flags(file, prot, flags, capabilities);
1203
David Howells8feae132009-01-08 12:04:47 +00001204 /* we're going to need to record the mapping */
1205 region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1206 if (!region)
1207 goto error_getting_region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208
David Howells8feae132009-01-08 12:04:47 +00001209 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1210 if (!vma)
1211 goto error_getting_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212
David Howells8feae132009-01-08 12:04:47 +00001213 atomic_set(&region->vm_usage, 1);
1214 region->vm_flags = vm_flags;
1215 region->vm_pgoff = pgoff;
1216
1217 INIT_LIST_HEAD(&vma->anon_vma_node);
1218 vma->vm_flags = vm_flags;
1219 vma->vm_pgoff = pgoff;
1220
1221 if (file) {
1222 region->vm_file = file;
1223 get_file(file);
1224 vma->vm_file = file;
1225 get_file(file);
1226 if (vm_flags & VM_EXECUTABLE) {
1227 added_exe_file_vma(current->mm);
1228 vma->vm_mm = current->mm;
1229 }
1230 }
1231
1232 down_write(&nommu_region_sem);
1233
1234 /* if we want to share, we need to check for regions created by other
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 * mmap() calls that overlap with our proposed mapping
David Howells8feae132009-01-08 12:04:47 +00001236 * - we can only share with a superset match on most regular files
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 * - shared mappings on character devices and memory backed files are
1238 * permitted to overlap inexactly as far as we are concerned for in
1239 * these cases, sharing is handled in the driver or filesystem rather
1240 * than here
1241 */
1242 if (vm_flags & VM_MAYSHARE) {
David Howells8feae132009-01-08 12:04:47 +00001243 struct vm_region *pregion;
1244 unsigned long pglen, rpglen, pgend, rpgend, start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245
David Howells8feae132009-01-08 12:04:47 +00001246 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1247 pgend = pgoff + pglen;
David Howells165b2392007-03-22 00:11:24 -08001248
David Howells8feae132009-01-08 12:04:47 +00001249 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1250 pregion = rb_entry(rb, struct vm_region, vm_rb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251
David Howells8feae132009-01-08 12:04:47 +00001252 if (!(pregion->vm_flags & VM_MAYSHARE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 continue;
1254
1255 /* search for overlapping mappings on the same file */
David Howells8feae132009-01-08 12:04:47 +00001256 if (pregion->vm_file->f_path.dentry->d_inode !=
1257 file->f_path.dentry->d_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258 continue;
1259
David Howells8feae132009-01-08 12:04:47 +00001260 if (pregion->vm_pgoff >= pgend)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 continue;
1262
David Howells8feae132009-01-08 12:04:47 +00001263 rpglen = pregion->vm_end - pregion->vm_start;
1264 rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1265 rpgend = pregion->vm_pgoff + rpglen;
1266 if (pgoff >= rpgend)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 continue;
1268
David Howells8feae132009-01-08 12:04:47 +00001269 /* handle inexactly overlapping matches between
1270 * mappings */
1271 if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1272 !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1273 /* new mapping is not a subset of the region */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 if (!(capabilities & BDI_CAP_MAP_DIRECT))
1275 goto sharing_violation;
1276 continue;
1277 }
1278
David Howells8feae132009-01-08 12:04:47 +00001279 /* we've found a region we can share */
1280 atomic_inc(&pregion->vm_usage);
1281 vma->vm_region = pregion;
1282 start = pregion->vm_start;
1283 start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1284 vma->vm_start = start;
1285 vma->vm_end = start + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286
David Howells8feae132009-01-08 12:04:47 +00001287 if (pregion->vm_flags & VM_MAPPED_COPY) {
1288 kdebug("share copy");
1289 vma->vm_flags |= VM_MAPPED_COPY;
1290 } else {
1291 kdebug("share mmap");
1292 ret = do_mmap_shared_file(vma);
1293 if (ret < 0) {
1294 vma->vm_region = NULL;
1295 vma->vm_start = 0;
1296 vma->vm_end = 0;
1297 atomic_dec(&pregion->vm_usage);
1298 pregion = NULL;
1299 goto error_just_free;
1300 }
1301 }
1302 fput(region->vm_file);
1303 kmem_cache_free(vm_region_jar, region);
1304 region = pregion;
1305 result = start;
1306 goto share;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 }
1308
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 /* obtain the address at which to make a shared mapping
1310 * - this is the hook for quasi-memory character devices to
1311 * tell us the location of a shared mapping
1312 */
1313 if (file && file->f_op->get_unmapped_area) {
1314 addr = file->f_op->get_unmapped_area(file, addr, len,
1315 pgoff, flags);
1316 if (IS_ERR((void *) addr)) {
1317 ret = addr;
1318 if (ret != (unsigned long) -ENOSYS)
David Howells8feae132009-01-08 12:04:47 +00001319 goto error_just_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320
1321 /* the driver refused to tell us where to site
1322 * the mapping so we'll have to attempt to copy
1323 * it */
1324 ret = (unsigned long) -ENODEV;
1325 if (!(capabilities & BDI_CAP_MAP_COPY))
David Howells8feae132009-01-08 12:04:47 +00001326 goto error_just_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327
1328 capabilities &= ~BDI_CAP_MAP_DIRECT;
David Howells8feae132009-01-08 12:04:47 +00001329 } else {
1330 vma->vm_start = region->vm_start = addr;
1331 vma->vm_end = region->vm_end = addr + len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 }
1333 }
1334 }
1335
David Howells8feae132009-01-08 12:04:47 +00001336 vma->vm_region = region;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337
1338 /* set up the mapping */
1339 if (file && vma->vm_flags & VM_SHARED)
David Howells8feae132009-01-08 12:04:47 +00001340 ret = do_mmap_shared_file(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 else
David Howells8feae132009-01-08 12:04:47 +00001342 ret = do_mmap_private(vma, region, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 if (ret < 0)
David Howells8feae132009-01-08 12:04:47 +00001344 goto error_put_region;
1345
1346 add_nommu_region(region);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347
1348 /* okay... we have a mapping; now we have to register it */
David Howells8feae132009-01-08 12:04:47 +00001349 result = vma->vm_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 current->mm->total_vm += len >> PAGE_SHIFT;
1352
David Howells8feae132009-01-08 12:04:47 +00001353share:
1354 add_vma_to_mm(current->mm, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355
David Howells8feae132009-01-08 12:04:47 +00001356 up_write(&nommu_region_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357
1358 if (prot & PROT_EXEC)
David Howells8feae132009-01-08 12:04:47 +00001359 flush_icache_range(result, result + len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360
David Howells8feae132009-01-08 12:04:47 +00001361 kleave(" = %lx", result);
1362 return result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363
David Howells8feae132009-01-08 12:04:47 +00001364error_put_region:
1365 __put_nommu_region(region);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 if (vma) {
Matt Helsley925d1c42008-04-29 01:01:36 -07001367 if (vma->vm_file) {
Gavin Lambert3fcd03e2006-09-30 23:27:01 -07001368 fput(vma->vm_file);
Matt Helsley925d1c42008-04-29 01:01:36 -07001369 if (vma->vm_flags & VM_EXECUTABLE)
1370 removed_exe_file_vma(vma->vm_mm);
1371 }
David Howells8feae132009-01-08 12:04:47 +00001372 kmem_cache_free(vm_area_cachep, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 }
David Howells8feae132009-01-08 12:04:47 +00001374 kleave(" = %d [pr]", ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375 return ret;
1376
David Howells8feae132009-01-08 12:04:47 +00001377error_just_free:
1378 up_write(&nommu_region_sem);
1379error:
1380 fput(region->vm_file);
1381 kmem_cache_free(vm_region_jar, region);
1382 fput(vma->vm_file);
1383 if (vma->vm_flags & VM_EXECUTABLE)
1384 removed_exe_file_vma(vma->vm_mm);
1385 kmem_cache_free(vm_area_cachep, vma);
1386 kleave(" = %d", ret);
1387 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388
David Howells8feae132009-01-08 12:04:47 +00001389sharing_violation:
1390 up_write(&nommu_region_sem);
1391 printk(KERN_WARNING "Attempt to share mismatched mappings\n");
1392 ret = -EINVAL;
1393 goto error;
1394
1395error_getting_vma:
1396 kmem_cache_free(vm_region_jar, region);
1397 printk(KERN_WARNING "Allocation of vma for %lu byte allocation"
1398 " from process %d failed\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 len, current->pid);
1400 show_free_areas();
1401 return -ENOMEM;
1402
David Howells8feae132009-01-08 12:04:47 +00001403error_getting_region:
1404 printk(KERN_WARNING "Allocation of vm region for %lu byte allocation"
1405 " from process %d failed\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 len, current->pid);
1407 show_free_areas();
1408 return -ENOMEM;
1409}
Paul Mundtb5073172007-07-21 04:37:25 -07001410EXPORT_SYMBOL(do_mmap_pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411
1412/*
David Howells8feae132009-01-08 12:04:47 +00001413 * split a vma into two pieces at address 'addr', a new vma is allocated either
1414 * for the first part or the tail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415 */
David Howells8feae132009-01-08 12:04:47 +00001416int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1417 unsigned long addr, int new_below)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418{
David Howells8feae132009-01-08 12:04:47 +00001419 struct vm_area_struct *new;
1420 struct vm_region *region;
1421 unsigned long npages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422
David Howells8feae132009-01-08 12:04:47 +00001423 kenter("");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424
David Howells8feae132009-01-08 12:04:47 +00001425 /* we're only permitted to split anonymous regions that have a single
1426 * owner */
1427 if (vma->vm_file ||
1428 atomic_read(&vma->vm_region->vm_usage) != 1)
1429 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430
David Howells8feae132009-01-08 12:04:47 +00001431 if (mm->map_count >= sysctl_max_map_count)
1432 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433
David Howells8feae132009-01-08 12:04:47 +00001434 region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1435 if (!region)
1436 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437
David Howells8feae132009-01-08 12:04:47 +00001438 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1439 if (!new) {
1440 kmem_cache_free(vm_region_jar, region);
1441 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442 }
David Howells8feae132009-01-08 12:04:47 +00001443
1444 /* most fields are the same, copy all, and then fixup */
1445 *new = *vma;
1446 *region = *vma->vm_region;
1447 new->vm_region = region;
1448
1449 npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1450
1451 if (new_below) {
Paul Mundtdd8632a2009-01-08 12:04:47 +00001452 region->vm_top = region->vm_end = new->vm_end = addr;
David Howells8feae132009-01-08 12:04:47 +00001453 } else {
1454 region->vm_start = new->vm_start = addr;
1455 region->vm_pgoff = new->vm_pgoff += npages;
1456 }
1457
1458 if (new->vm_ops && new->vm_ops->open)
1459 new->vm_ops->open(new);
1460
1461 delete_vma_from_mm(vma);
1462 down_write(&nommu_region_sem);
1463 delete_nommu_region(vma->vm_region);
1464 if (new_below) {
1465 vma->vm_region->vm_start = vma->vm_start = addr;
1466 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1467 } else {
1468 vma->vm_region->vm_end = vma->vm_end = addr;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001469 vma->vm_region->vm_top = addr;
David Howells8feae132009-01-08 12:04:47 +00001470 }
1471 add_nommu_region(vma->vm_region);
1472 add_nommu_region(new->vm_region);
1473 up_write(&nommu_region_sem);
1474 add_vma_to_mm(mm, vma);
1475 add_vma_to_mm(mm, new);
1476 return 0;
1477}
1478
1479/*
1480 * shrink a VMA by removing the specified chunk from either the beginning or
1481 * the end
1482 */
1483static int shrink_vma(struct mm_struct *mm,
1484 struct vm_area_struct *vma,
1485 unsigned long from, unsigned long to)
1486{
1487 struct vm_region *region;
1488
1489 kenter("");
1490
1491 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1492 * and list */
1493 delete_vma_from_mm(vma);
1494 if (from > vma->vm_start)
1495 vma->vm_end = from;
1496 else
1497 vma->vm_start = to;
1498 add_vma_to_mm(mm, vma);
1499
1500 /* cut the backing region down to size */
1501 region = vma->vm_region;
1502 BUG_ON(atomic_read(&region->vm_usage) != 1);
1503
1504 down_write(&nommu_region_sem);
1505 delete_nommu_region(region);
Paul Mundtdd8632a2009-01-08 12:04:47 +00001506 if (from > region->vm_start) {
1507 to = region->vm_top;
1508 region->vm_top = region->vm_end = from;
1509 } else {
David Howells8feae132009-01-08 12:04:47 +00001510 region->vm_start = to;
Paul Mundtdd8632a2009-01-08 12:04:47 +00001511 }
David Howells8feae132009-01-08 12:04:47 +00001512 add_nommu_region(region);
1513 up_write(&nommu_region_sem);
1514
1515 free_page_series(from, to);
1516 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517}
1518
David Howells30340972006-09-27 01:50:20 -07001519/*
1520 * release a mapping
David Howells8feae132009-01-08 12:04:47 +00001521 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1522 * VMA, though it need not cover the whole VMA
David Howells30340972006-09-27 01:50:20 -07001523 */
David Howells8feae132009-01-08 12:04:47 +00001524int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525{
David Howells8feae132009-01-08 12:04:47 +00001526 struct vm_area_struct *vma;
1527 struct rb_node *rb;
1528 unsigned long end = start + len;
1529 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530
David Howells8feae132009-01-08 12:04:47 +00001531 kenter(",%lx,%zx", start, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532
David Howells8feae132009-01-08 12:04:47 +00001533 if (len == 0)
1534 return -EINVAL;
1535
1536 /* find the first potentially overlapping VMA */
1537 vma = find_vma(mm, start);
1538 if (!vma) {
1539 printk(KERN_WARNING
1540 "munmap of memory not mmapped by process %d (%s):"
1541 " 0x%lx-0x%lx\n",
1542 current->pid, current->comm, start, start + len - 1);
1543 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
Simon Derr2f60f8d2005-08-04 19:52:03 -07001852 /*
1853 * cast `allowed' as a signed long because vm_committed_space
1854 * sometimes has a negative value
1855 */
Alan Cox80119ef2008-05-23 13:04:31 -07001856 if (atomic_long_read(&vm_committed_space) < (long)allowed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857 return 0;
Hideo AOKId5ddc792006-04-10 22:53:01 -07001858error:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001859 vm_unacct_memory(pages);
1860
1861 return -ENOMEM;
1862}
1863
1864int in_gate_area_no_task(unsigned long addr)
1865{
1866 return 0;
1867}
David Howellsb0e15192006-01-06 00:11:42 -08001868
Nick Piggind0217ac2007-07-19 01:47:03 -07001869int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
David Howellsb0e15192006-01-06 00:11:42 -08001870{
1871 BUG();
Nick Piggind0217ac2007-07-19 01:47:03 -07001872 return 0;
David Howellsb0e15192006-01-06 00:11:42 -08001873}
Paul Mundtb5073172007-07-21 04:37:25 -07001874EXPORT_SYMBOL(filemap_fault);
David Howells0ec76a12006-09-27 01:50:15 -07001875
1876/*
1877 * Access another process' address space.
1878 * - source/target buffer must be kernel space
1879 */
1880int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
1881{
David Howells0ec76a12006-09-27 01:50:15 -07001882 struct vm_area_struct *vma;
1883 struct mm_struct *mm;
1884
1885 if (addr + len < addr)
1886 return 0;
1887
1888 mm = get_task_mm(tsk);
1889 if (!mm)
1890 return 0;
1891
1892 down_read(&mm->mmap_sem);
1893
1894 /* the access must start within one of the target process's mappings */
David Howells0159b142006-09-27 01:50:16 -07001895 vma = find_vma(mm, addr);
1896 if (vma) {
David Howells0ec76a12006-09-27 01:50:15 -07001897 /* don't overrun this mapping */
1898 if (addr + len >= vma->vm_end)
1899 len = vma->vm_end - addr;
1900
1901 /* only read or write mappings where it is permitted */
David Howellsd00c7b992006-09-27 01:50:19 -07001902 if (write && vma->vm_flags & VM_MAYWRITE)
David Howells0ec76a12006-09-27 01:50:15 -07001903 len -= copy_to_user((void *) addr, buf, len);
David Howellsd00c7b992006-09-27 01:50:19 -07001904 else if (!write && vma->vm_flags & VM_MAYREAD)
David Howells0ec76a12006-09-27 01:50:15 -07001905 len -= copy_from_user(buf, (void *) addr, len);
1906 else
1907 len = 0;
1908 } else {
1909 len = 0;
1910 }
1911
1912 up_read(&mm->mmap_sem);
1913 mmput(mm);
1914 return len;
1915}