blob: 7695dc850785e853f066e63992c242ffee14c30c [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 *
9 * Copyright (c) 2004-2005 David Howells <dhowells@redhat.com>
10 * 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 Mundtf905bc42008-02-04 22:29:59 -080013 * Copyright (c) 2007 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>
36
Nick Pigginb291f002008-10-18 20:26:44 -070037#include "internal.h"
38
Linus Torvalds1da177e2005-04-16 15:20:36 -070039void *high_memory;
40struct page *mem_map;
41unsigned long max_mapnr;
42unsigned long num_physpages;
43unsigned long askedalloc, realalloc;
Alan Cox80119ef2008-05-23 13:04:31 -070044atomic_long_t vm_committed_space = ATOMIC_LONG_INIT(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -070045int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
46int sysctl_overcommit_ratio = 50; /* default is 50% */
47int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
48int heap_stack_gap = 0;
49
50EXPORT_SYMBOL(mem_map);
Wu, Bryan6a04de62007-04-11 23:28:47 -070051EXPORT_SYMBOL(num_physpages);
Linus Torvalds1da177e2005-04-16 15:20:36 -070052
53/* list of shareable VMAs */
54struct rb_root nommu_vma_tree = RB_ROOT;
55DECLARE_RWSEM(nommu_vma_sem);
56
57struct vm_operations_struct generic_file_vm_ops = {
58};
59
60/*
61 * Handle all mappings that got truncated by a "truncate()"
62 * system call.
63 *
64 * NOTE! We have to be ready to update the memory sharing
65 * between the file and the memory map for a potential last
66 * incomplete page. Ugly, but necessary.
67 */
68int vmtruncate(struct inode *inode, loff_t offset)
69{
70 struct address_space *mapping = inode->i_mapping;
71 unsigned long limit;
72
73 if (inode->i_size < offset)
74 goto do_expand;
75 i_size_write(inode, offset);
76
77 truncate_inode_pages(mapping, offset);
78 goto out_truncate;
79
80do_expand:
81 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
82 if (limit != RLIM_INFINITY && offset > limit)
83 goto out_sig;
84 if (offset > inode->i_sb->s_maxbytes)
85 goto out;
86 i_size_write(inode, offset);
87
88out_truncate:
89 if (inode->i_op && inode->i_op->truncate)
90 inode->i_op->truncate(inode);
91 return 0;
92out_sig:
93 send_sig(SIGXFSZ, current, 0);
94out:
95 return -EFBIG;
96}
97
98EXPORT_SYMBOL(vmtruncate);
99
100/*
101 * Return the total memory allocated for this pointer, not
102 * just what the caller asked for.
103 *
104 * Doesn't have to be accurate, i.e. may have races.
105 */
106unsigned int kobjsize(const void *objp)
107{
108 struct page *page;
109
Michael Hennerich4016a132008-04-28 02:13:38 -0700110 /*
111 * If the object we have should not have ksize performed on it,
112 * return size of 0
113 */
Paul Mundt5a1603b2008-06-12 16:29:55 +0900114 if (!objp || !virt_addr_valid(objp))
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700115 return 0;
116
117 page = virt_to_head_page(objp);
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700118
119 /*
120 * If the allocator sets PageSlab, we know the pointer came from
121 * kmalloc().
122 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123 if (PageSlab(page))
124 return ksize(objp);
125
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700126 /*
127 * The ksize() function is only guaranteed to work for pointers
Paul Mundt5a1603b2008-06-12 16:29:55 +0900128 * returned by kmalloc(). So handle arbitrary pointers here.
Paul Mundt6cfd53fc2008-06-05 22:46:08 -0700129 */
Paul Mundt5a1603b2008-06-12 16:29:55 +0900130 return PAGE_SIZE << compound_order(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131}
132
Nick Pigginb291f002008-10-18 20:26:44 -0700133int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
134 unsigned long start, int len, int flags,
135 struct page **pages, struct vm_area_struct **vmas)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136{
Sonic Zhang910e46d2006-09-27 01:50:17 -0700137 struct vm_area_struct *vma;
David Howells7b4d5b82006-09-27 01:50:18 -0700138 unsigned long vm_flags;
139 int i;
Nick Pigginb291f002008-10-18 20:26:44 -0700140 int write = !!(flags & GUP_FLAGS_WRITE);
141 int force = !!(flags & GUP_FLAGS_FORCE);
142 int ignore = !!(flags & GUP_FLAGS_IGNORE_VMA_PERMISSIONS);
David Howells7b4d5b82006-09-27 01:50:18 -0700143
144 /* calculate required read or write permissions.
145 * - if 'force' is set, we only require the "MAY" flags.
146 */
147 vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
148 vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149
150 for (i = 0; i < len; i++) {
Sonic Zhang910e46d2006-09-27 01:50:17 -0700151 vma = find_vma(mm, start);
David Howells7b4d5b82006-09-27 01:50:18 -0700152 if (!vma)
153 goto finish_or_fault;
154
155 /* protect what we can, including chardevs */
156 if (vma->vm_flags & (VM_IO | VM_PFNMAP) ||
Nick Pigginb291f002008-10-18 20:26:44 -0700157 (!ignore && !(vm_flags & vma->vm_flags)))
David Howells7b4d5b82006-09-27 01:50:18 -0700158 goto finish_or_fault;
Sonic Zhang910e46d2006-09-27 01:50:17 -0700159
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160 if (pages) {
161 pages[i] = virt_to_page(start);
162 if (pages[i])
163 page_cache_get(pages[i]);
164 }
165 if (vmas)
Sonic Zhang910e46d2006-09-27 01:50:17 -0700166 vmas[i] = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167 start += PAGE_SIZE;
168 }
David Howells7b4d5b82006-09-27 01:50:18 -0700169
170 return i;
171
172finish_or_fault:
173 return i ? : -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174}
Nick Pigginb291f002008-10-18 20:26:44 -0700175
176
177/*
178 * get a list of pages in an address range belonging to the specified process
179 * and indicate the VMA that covers each page
180 * - this is potentially dodgy as we may end incrementing the page count of a
181 * slab page or a secondary page from a compound page
182 * - don't permit access to VMAs that don't support it, such as I/O mappings
183 */
184int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
185 unsigned long start, int len, int write, int force,
186 struct page **pages, struct vm_area_struct **vmas)
187{
188 int flags = 0;
189
190 if (write)
191 flags |= GUP_FLAGS_WRITE;
192 if (force)
193 flags |= GUP_FLAGS_FORCE;
194
195 return __get_user_pages(tsk, mm,
196 start, len, flags,
197 pages, vmas);
198}
Greg Ungerer66aa2b42005-09-12 11:18:10 +1000199EXPORT_SYMBOL(get_user_pages);
200
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201DEFINE_RWLOCK(vmlist_lock);
202struct vm_struct *vmlist;
203
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800204void vfree(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205{
206 kfree(addr);
207}
Paul Mundtb5073172007-07-21 04:37:25 -0700208EXPORT_SYMBOL(vfree);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209
Al Virodd0fc662005-10-07 07:46:04 +0100210void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211{
212 /*
Robert P. J. Day85186092007-10-19 23:11:38 +0200213 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
214 * returns only a logical address.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 */
Nick Piggin84097512006-03-22 00:08:34 -0800216 return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217}
Paul Mundtb5073172007-07-21 04:37:25 -0700218EXPORT_SYMBOL(__vmalloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219
Paul Mundtf905bc42008-02-04 22:29:59 -0800220void *vmalloc_user(unsigned long size)
221{
222 void *ret;
223
224 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
225 PAGE_KERNEL);
226 if (ret) {
227 struct vm_area_struct *vma;
228
229 down_write(&current->mm->mmap_sem);
230 vma = find_vma(current->mm, (unsigned long)ret);
231 if (vma)
232 vma->vm_flags |= VM_USERMAP;
233 up_write(&current->mm->mmap_sem);
234 }
235
236 return ret;
237}
238EXPORT_SYMBOL(vmalloc_user);
239
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800240struct page *vmalloc_to_page(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241{
242 return virt_to_page(addr);
243}
Paul Mundtb5073172007-07-21 04:37:25 -0700244EXPORT_SYMBOL(vmalloc_to_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800246unsigned long vmalloc_to_pfn(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247{
248 return page_to_pfn(virt_to_page(addr));
249}
Paul Mundtb5073172007-07-21 04:37:25 -0700250EXPORT_SYMBOL(vmalloc_to_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251
252long vread(char *buf, char *addr, unsigned long count)
253{
254 memcpy(buf, addr, count);
255 return count;
256}
257
258long vwrite(char *buf, char *addr, unsigned long count)
259{
260 /* Don't allow overflow */
261 if ((unsigned long) addr + count < count)
262 count = -(unsigned long) addr;
263
264 memcpy(addr, buf, count);
265 return(count);
266}
267
268/*
269 * vmalloc - allocate virtually continguos memory
270 *
271 * @size: allocation size
272 *
273 * Allocate enough pages to cover @size from the page level
274 * allocator and map them into continguos kernel virtual space.
275 *
Michael Opdenackerc1c88972006-10-03 23:21:02 +0200276 * For tight control over page level allocator and protection flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 * use __vmalloc() instead.
278 */
279void *vmalloc(unsigned long size)
280{
281 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
282}
Andrew Mortonf6138882006-02-28 16:59:18 -0800283EXPORT_SYMBOL(vmalloc);
284
285void *vmalloc_node(unsigned long size, int node)
286{
287 return vmalloc(size);
288}
289EXPORT_SYMBOL(vmalloc_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290
Paul Mundt1af446e2008-08-04 16:01:47 +0900291#ifndef PAGE_KERNEL_EXEC
292# define PAGE_KERNEL_EXEC PAGE_KERNEL
293#endif
294
295/**
296 * vmalloc_exec - allocate virtually contiguous, executable memory
297 * @size: allocation size
298 *
299 * Kernel-internal function to allocate enough pages to cover @size
300 * the page level allocator and map them into contiguous and
301 * executable kernel virtual space.
302 *
303 * For tight control over page level allocator and protection flags
304 * use __vmalloc() instead.
305 */
306
307void *vmalloc_exec(unsigned long size)
308{
309 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
310}
311
Paul Mundtb5073172007-07-21 04:37:25 -0700312/**
313 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314 * @size: allocation size
315 *
316 * Allocate enough 32bit PA addressable pages to cover @size from the
317 * page level allocator and map them into continguos kernel virtual space.
318 */
319void *vmalloc_32(unsigned long size)
320{
321 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
322}
Paul Mundtb5073172007-07-21 04:37:25 -0700323EXPORT_SYMBOL(vmalloc_32);
324
325/**
326 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
327 * @size: allocation size
328 *
329 * The resulting memory area is 32bit addressable and zeroed so it can be
330 * mapped to userspace without leaking data.
Paul Mundtf905bc42008-02-04 22:29:59 -0800331 *
332 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
333 * remap_vmalloc_range() are permissible.
Paul Mundtb5073172007-07-21 04:37:25 -0700334 */
335void *vmalloc_32_user(unsigned long size)
336{
Paul Mundtf905bc42008-02-04 22:29:59 -0800337 /*
338 * We'll have to sort out the ZONE_DMA bits for 64-bit,
339 * but for now this can simply use vmalloc_user() directly.
340 */
341 return vmalloc_user(size);
Paul Mundtb5073172007-07-21 04:37:25 -0700342}
343EXPORT_SYMBOL(vmalloc_32_user);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344
345void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
346{
347 BUG();
348 return NULL;
349}
Paul Mundtb5073172007-07-21 04:37:25 -0700350EXPORT_SYMBOL(vmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351
Christoph Lameterb3bdda02008-02-04 22:28:32 -0800352void vunmap(const void *addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353{
354 BUG();
355}
Paul Mundtb5073172007-07-21 04:37:25 -0700356EXPORT_SYMBOL(vunmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357
358/*
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -0700359 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
360 * have one.
361 */
362void __attribute__((weak)) vmalloc_sync_all(void)
363{
364}
365
Paul Mundtb5073172007-07-21 04:37:25 -0700366int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
367 struct page *page)
368{
369 return -EINVAL;
370}
371EXPORT_SYMBOL(vm_insert_page);
372
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -0700373/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 * sys_brk() for the most part doesn't need the global kernel
375 * lock, except when an application is doing something nasty
376 * like trying to un-brk an area that has already been mapped
377 * to a regular file. in this case, the unmapping will need
378 * to invoke file system routines that need the global lock.
379 */
380asmlinkage unsigned long sys_brk(unsigned long brk)
381{
382 struct mm_struct *mm = current->mm;
383
384 if (brk < mm->start_brk || brk > mm->context.end_brk)
385 return mm->brk;
386
387 if (mm->brk == brk)
388 return mm->brk;
389
390 /*
391 * Always allow shrinking brk
392 */
393 if (brk <= mm->brk) {
394 mm->brk = brk;
395 return brk;
396 }
397
398 /*
399 * Ok, looks good - let it rip.
400 */
401 return mm->brk = brk;
402}
403
404#ifdef DEBUG
405static void show_process_blocks(void)
406{
407 struct vm_list_struct *vml;
408
409 printk("Process blocks %d:", current->pid);
410
411 for (vml = &current->mm->context.vmlist; vml; vml = vml->next) {
412 printk(" %p: %p", vml, vml->vma);
413 if (vml->vma)
414 printk(" (%d @%lx #%d)",
415 kobjsize((void *) vml->vma->vm_start),
416 vml->vma->vm_start,
417 atomic_read(&vml->vma->vm_usage));
418 printk(vml->next ? " ->" : ".\n");
419 }
420}
421#endif /* DEBUG */
422
David Howells30340972006-09-27 01:50:20 -0700423/*
424 * add a VMA into a process's mm_struct in the appropriate place in the list
425 * - should be called with mm->mmap_sem held writelocked
426 */
427static void add_vma_to_mm(struct mm_struct *mm, struct vm_list_struct *vml)
428{
429 struct vm_list_struct **ppv;
430
431 for (ppv = &current->mm->context.vmlist; *ppv; ppv = &(*ppv)->next)
432 if ((*ppv)->vma->vm_start > vml->vma->vm_start)
433 break;
434
435 vml->next = *ppv;
436 *ppv = vml;
437}
438
439/*
440 * look up the first VMA in which addr resides, NULL if none
441 * - should be called with mm->mmap_sem at least held readlocked
442 */
443struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
444{
445 struct vm_list_struct *loop, *vml;
446
447 /* search the vm_start ordered list */
448 vml = NULL;
449 for (loop = mm->context.vmlist; loop; loop = loop->next) {
450 if (loop->vma->vm_start > addr)
451 break;
452 vml = loop;
453 }
454
455 if (vml && vml->vma->vm_end > addr)
456 return vml->vma;
457
458 return NULL;
459}
460EXPORT_SYMBOL(find_vma);
461
462/*
David Howells930e6522006-09-27 01:50:22 -0700463 * find a VMA
464 * - we don't extend stack VMAs under NOMMU conditions
465 */
466struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
467{
468 return find_vma(mm, addr);
469}
470
Greg Ungerer57c8f632007-07-15 23:38:28 -0700471int expand_stack(struct vm_area_struct *vma, unsigned long address)
472{
473 return -ENOMEM;
474}
475
David Howells930e6522006-09-27 01:50:22 -0700476/*
David Howells6fa5f802006-09-27 01:50:21 -0700477 * look up the first VMA exactly that exactly matches addr
478 * - should be called with mm->mmap_sem at least held readlocked
479 */
480static inline struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
481 unsigned long addr)
482{
483 struct vm_list_struct *vml;
484
485 /* search the vm_start ordered list */
486 for (vml = mm->context.vmlist; vml; vml = vml->next) {
487 if (vml->vma->vm_start == addr)
488 return vml->vma;
489 if (vml->vma->vm_start > addr)
490 break;
491 }
492
493 return NULL;
494}
495
496/*
David Howells30340972006-09-27 01:50:20 -0700497 * find a VMA in the global tree
498 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499static inline struct vm_area_struct *find_nommu_vma(unsigned long start)
500{
501 struct vm_area_struct *vma;
502 struct rb_node *n = nommu_vma_tree.rb_node;
503
504 while (n) {
505 vma = rb_entry(n, struct vm_area_struct, vm_rb);
506
507 if (start < vma->vm_start)
508 n = n->rb_left;
509 else if (start > vma->vm_start)
510 n = n->rb_right;
511 else
512 return vma;
513 }
514
515 return NULL;
516}
517
David Howells30340972006-09-27 01:50:20 -0700518/*
519 * add a VMA in the global tree
520 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521static void add_nommu_vma(struct vm_area_struct *vma)
522{
523 struct vm_area_struct *pvma;
524 struct address_space *mapping;
525 struct rb_node **p = &nommu_vma_tree.rb_node;
526 struct rb_node *parent = NULL;
527
528 /* add the VMA to the mapping */
529 if (vma->vm_file) {
530 mapping = vma->vm_file->f_mapping;
531
532 flush_dcache_mmap_lock(mapping);
533 vma_prio_tree_insert(vma, &mapping->i_mmap);
534 flush_dcache_mmap_unlock(mapping);
535 }
536
537 /* add the VMA to the master list */
538 while (*p) {
539 parent = *p;
540 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
541
542 if (vma->vm_start < pvma->vm_start) {
543 p = &(*p)->rb_left;
544 }
545 else if (vma->vm_start > pvma->vm_start) {
546 p = &(*p)->rb_right;
547 }
548 else {
549 /* mappings are at the same address - this can only
550 * happen for shared-mem chardevs and shared file
551 * mappings backed by ramfs/tmpfs */
552 BUG_ON(!(pvma->vm_flags & VM_SHARED));
553
554 if (vma < pvma)
555 p = &(*p)->rb_left;
556 else if (vma > pvma)
557 p = &(*p)->rb_right;
558 else
559 BUG();
560 }
561 }
562
563 rb_link_node(&vma->vm_rb, parent, p);
564 rb_insert_color(&vma->vm_rb, &nommu_vma_tree);
565}
566
David Howells30340972006-09-27 01:50:20 -0700567/*
568 * delete a VMA from the global list
569 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570static void delete_nommu_vma(struct vm_area_struct *vma)
571{
572 struct address_space *mapping;
573
574 /* remove the VMA from the mapping */
575 if (vma->vm_file) {
576 mapping = vma->vm_file->f_mapping;
577
578 flush_dcache_mmap_lock(mapping);
579 vma_prio_tree_remove(vma, &mapping->i_mmap);
580 flush_dcache_mmap_unlock(mapping);
581 }
582
583 /* remove from the master list */
584 rb_erase(&vma->vm_rb, &nommu_vma_tree);
585}
586
587/*
588 * determine whether a mapping should be permitted and, if so, what sort of
589 * mapping we're capable of supporting
590 */
591static int validate_mmap_request(struct file *file,
592 unsigned long addr,
593 unsigned long len,
594 unsigned long prot,
595 unsigned long flags,
596 unsigned long pgoff,
597 unsigned long *_capabilities)
598{
599 unsigned long capabilities;
600 unsigned long reqprot = prot;
601 int ret;
602
603 /* do the simple checks first */
604 if (flags & MAP_FIXED || addr) {
605 printk(KERN_DEBUG
606 "%d: Can't do fixed-address/overlay mmap of RAM\n",
607 current->pid);
608 return -EINVAL;
609 }
610
611 if ((flags & MAP_TYPE) != MAP_PRIVATE &&
612 (flags & MAP_TYPE) != MAP_SHARED)
613 return -EINVAL;
614
Mike Frysingerf81cff02006-12-06 12:02:59 +1000615 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 return -EINVAL;
617
Mike Frysingerf81cff02006-12-06 12:02:59 +1000618 /* Careful about overflows.. */
619 len = PAGE_ALIGN(len);
620 if (!len || len > TASK_SIZE)
621 return -ENOMEM;
622
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 /* offset overflow? */
624 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
Mike Frysingerf81cff02006-12-06 12:02:59 +1000625 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626
627 if (file) {
628 /* validate file mapping requests */
629 struct address_space *mapping;
630
631 /* files must support mmap */
632 if (!file->f_op || !file->f_op->mmap)
633 return -ENODEV;
634
635 /* work out if what we've got could possibly be shared
636 * - we support chardevs that provide their own "memory"
637 * - we support files/blockdevs that are memory backed
638 */
639 mapping = file->f_mapping;
640 if (!mapping)
Josef Sipeke9536ae2006-12-08 02:37:21 -0800641 mapping = file->f_path.dentry->d_inode->i_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642
643 capabilities = 0;
644 if (mapping && mapping->backing_dev_info)
645 capabilities = mapping->backing_dev_info->capabilities;
646
647 if (!capabilities) {
648 /* no explicit capabilities set, so assume some
649 * defaults */
Josef Sipeke9536ae2006-12-08 02:37:21 -0800650 switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 case S_IFREG:
652 case S_IFBLK:
653 capabilities = BDI_CAP_MAP_COPY;
654 break;
655
656 case S_IFCHR:
657 capabilities =
658 BDI_CAP_MAP_DIRECT |
659 BDI_CAP_READ_MAP |
660 BDI_CAP_WRITE_MAP;
661 break;
662
663 default:
664 return -EINVAL;
665 }
666 }
667
668 /* eliminate any capabilities that we can't support on this
669 * device */
670 if (!file->f_op->get_unmapped_area)
671 capabilities &= ~BDI_CAP_MAP_DIRECT;
672 if (!file->f_op->read)
673 capabilities &= ~BDI_CAP_MAP_COPY;
674
675 if (flags & MAP_SHARED) {
676 /* do checks for writing, appending and locking */
677 if ((prot & PROT_WRITE) &&
678 !(file->f_mode & FMODE_WRITE))
679 return -EACCES;
680
Josef Sipeke9536ae2006-12-08 02:37:21 -0800681 if (IS_APPEND(file->f_path.dentry->d_inode) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 (file->f_mode & FMODE_WRITE))
683 return -EACCES;
684
Josef Sipeke9536ae2006-12-08 02:37:21 -0800685 if (locks_verify_locked(file->f_path.dentry->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 return -EAGAIN;
687
688 if (!(capabilities & BDI_CAP_MAP_DIRECT))
689 return -ENODEV;
690
691 if (((prot & PROT_READ) && !(capabilities & BDI_CAP_READ_MAP)) ||
692 ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
693 ((prot & PROT_EXEC) && !(capabilities & BDI_CAP_EXEC_MAP))
694 ) {
695 printk("MAP_SHARED not completely supported on !MMU\n");
696 return -EINVAL;
697 }
698
699 /* we mustn't privatise shared mappings */
700 capabilities &= ~BDI_CAP_MAP_COPY;
701 }
702 else {
703 /* we're going to read the file into private memory we
704 * allocate */
705 if (!(capabilities & BDI_CAP_MAP_COPY))
706 return -ENODEV;
707
708 /* we don't permit a private writable mapping to be
709 * shared with the backing device */
710 if (prot & PROT_WRITE)
711 capabilities &= ~BDI_CAP_MAP_DIRECT;
712 }
713
714 /* handle executable mappings and implied executable
715 * mappings */
Josef Sipeke9536ae2006-12-08 02:37:21 -0800716 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 if (prot & PROT_EXEC)
718 return -EPERM;
719 }
720 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
721 /* handle implication of PROT_EXEC by PROT_READ */
722 if (current->personality & READ_IMPLIES_EXEC) {
723 if (capabilities & BDI_CAP_EXEC_MAP)
724 prot |= PROT_EXEC;
725 }
726 }
727 else if ((prot & PROT_READ) &&
728 (prot & PROT_EXEC) &&
729 !(capabilities & BDI_CAP_EXEC_MAP)
730 ) {
731 /* backing file is not executable, try to copy */
732 capabilities &= ~BDI_CAP_MAP_DIRECT;
733 }
734 }
735 else {
736 /* anonymous mappings are always memory backed and can be
737 * privately mapped
738 */
739 capabilities = BDI_CAP_MAP_COPY;
740
741 /* handle PROT_EXEC implication by PROT_READ */
742 if ((prot & PROT_READ) &&
743 (current->personality & READ_IMPLIES_EXEC))
744 prot |= PROT_EXEC;
745 }
746
747 /* allow the security API to have its say */
Eric Parised032182007-06-28 15:55:21 -0400748 ret = security_file_mmap(file, reqprot, prot, flags, addr, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749 if (ret < 0)
750 return ret;
751
752 /* looks okay */
753 *_capabilities = capabilities;
754 return 0;
755}
756
757/*
758 * we've determined that we can make the mapping, now translate what we
759 * now know into VMA flags
760 */
761static unsigned long determine_vm_flags(struct file *file,
762 unsigned long prot,
763 unsigned long flags,
764 unsigned long capabilities)
765{
766 unsigned long vm_flags;
767
768 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
769 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
770 /* vm_flags |= mm->def_flags; */
771
772 if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
773 /* attempt to share read-only copies of mapped file chunks */
774 if (file && !(prot & PROT_WRITE))
775 vm_flags |= VM_MAYSHARE;
776 }
777 else {
778 /* overlay a shareable mapping on the backing device or inode
779 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
780 * romfs/cramfs */
781 if (flags & MAP_SHARED)
782 vm_flags |= VM_MAYSHARE | VM_SHARED;
783 else if ((((vm_flags & capabilities) ^ vm_flags) & BDI_CAP_VMFLAGS) == 0)
784 vm_flags |= VM_MAYSHARE;
785 }
786
787 /* refuse to let anyone share private mappings with this process if
788 * it's being traced - otherwise breakpoints set in it may interfere
789 * with another untraced process
790 */
Roland McGrathfa8e26c2008-07-25 19:45:50 -0700791 if ((flags & MAP_PRIVATE) && tracehook_expect_breakpoints(current))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 vm_flags &= ~VM_MAYSHARE;
793
794 return vm_flags;
795}
796
797/*
798 * set up a shared mapping on a file
799 */
800static int do_mmap_shared_file(struct vm_area_struct *vma, unsigned long len)
801{
802 int ret;
803
804 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
805 if (ret != -ENOSYS)
806 return ret;
807
808 /* getting an ENOSYS error indicates that direct mmap isn't
809 * possible (as opposed to tried but failed) so we'll fall
810 * through to making a private copy of the data and mapping
811 * that if we can */
812 return -ENODEV;
813}
814
815/*
816 * set up a private mapping or an anonymous shared mapping
817 */
818static int do_mmap_private(struct vm_area_struct *vma, unsigned long len)
819{
820 void *base;
821 int ret;
822
823 /* invoke the file's mapping function so that it can keep track of
824 * shared mappings on devices or memory
825 * - VM_MAYSHARE will be set if it may attempt to share
826 */
827 if (vma->vm_file) {
828 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
829 if (ret != -ENOSYS) {
830 /* shouldn't return success if we're not sharing */
831 BUG_ON(ret == 0 && !(vma->vm_flags & VM_MAYSHARE));
832 return ret; /* success or a real error */
833 }
834
835 /* getting an ENOSYS error indicates that direct mmap isn't
836 * possible (as opposed to tried but failed) so we'll try to
837 * make a private copy of the data and map that instead */
838 }
839
840 /* allocate some memory to hold the mapping
841 * - note that this may not return a page-aligned address if the object
842 * we're allocating is smaller than a page
843 */
Nick Piggin84097512006-03-22 00:08:34 -0800844 base = kmalloc(len, GFP_KERNEL|__GFP_COMP);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 if (!base)
846 goto enomem;
847
848 vma->vm_start = (unsigned long) base;
849 vma->vm_end = vma->vm_start + len;
850 vma->vm_flags |= VM_MAPPED_COPY;
851
852#ifdef WARN_ON_SLACK
853 if (len + WARN_ON_SLACK <= kobjsize(result))
854 printk("Allocation of %lu bytes from process %d has %lu bytes of slack\n",
855 len, current->pid, kobjsize(result) - len);
856#endif
857
858 if (vma->vm_file) {
859 /* read the contents of a file into the copy */
860 mm_segment_t old_fs;
861 loff_t fpos;
862
863 fpos = vma->vm_pgoff;
864 fpos <<= PAGE_SHIFT;
865
866 old_fs = get_fs();
867 set_fs(KERNEL_DS);
868 ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos);
869 set_fs(old_fs);
870
871 if (ret < 0)
872 goto error_free;
873
874 /* clear the last little bit */
875 if (ret < len)
876 memset(base + ret, 0, len - ret);
877
878 } else {
879 /* if it's an anonymous mapping, then just clear it */
880 memset(base, 0, len);
881 }
882
883 return 0;
884
885error_free:
886 kfree(base);
887 vma->vm_start = 0;
888 return ret;
889
890enomem:
891 printk("Allocation of length %lu from process %d failed\n",
892 len, current->pid);
893 show_free_areas();
894 return -ENOMEM;
895}
896
897/*
898 * handle mapping creation for uClinux
899 */
900unsigned long do_mmap_pgoff(struct file *file,
901 unsigned long addr,
902 unsigned long len,
903 unsigned long prot,
904 unsigned long flags,
905 unsigned long pgoff)
906{
907 struct vm_list_struct *vml = NULL;
908 struct vm_area_struct *vma = NULL;
909 struct rb_node *rb;
910 unsigned long capabilities, vm_flags;
911 void *result;
912 int ret;
913
Eric Paris7cd94142007-11-26 18:47:40 -0500914 if (!(flags & MAP_FIXED))
915 addr = round_hint_to_min(addr);
916
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 /* decide whether we should attempt the mapping, and if so what sort of
918 * mapping */
919 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
920 &capabilities);
921 if (ret < 0)
922 return ret;
923
924 /* we've determined that we can make the mapping, now translate what we
925 * now know into VMA flags */
926 vm_flags = determine_vm_flags(file, prot, flags, capabilities);
927
928 /* we're going to need to record the mapping if it works */
Burman Yan4668edc2006-12-06 20:38:51 -0800929 vml = kzalloc(sizeof(struct vm_list_struct), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 if (!vml)
931 goto error_getting_vml;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932
933 down_write(&nommu_vma_sem);
934
935 /* if we want to share, we need to check for VMAs created by other
936 * mmap() calls that overlap with our proposed mapping
937 * - we can only share with an exact match on most regular files
938 * - shared mappings on character devices and memory backed files are
939 * permitted to overlap inexactly as far as we are concerned for in
940 * these cases, sharing is handled in the driver or filesystem rather
941 * than here
942 */
943 if (vm_flags & VM_MAYSHARE) {
944 unsigned long pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
945 unsigned long vmpglen;
946
David Howells165b2392007-03-22 00:11:24 -0800947 /* suppress VMA sharing for shared regions */
948 if (vm_flags & VM_SHARED &&
949 capabilities & BDI_CAP_MAP_DIRECT)
950 goto dont_share_VMAs;
951
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 for (rb = rb_first(&nommu_vma_tree); rb; rb = rb_next(rb)) {
953 vma = rb_entry(rb, struct vm_area_struct, vm_rb);
954
955 if (!(vma->vm_flags & VM_MAYSHARE))
956 continue;
957
958 /* search for overlapping mappings on the same file */
Josef Sipeke9536ae2006-12-08 02:37:21 -0800959 if (vma->vm_file->f_path.dentry->d_inode != file->f_path.dentry->d_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 continue;
961
962 if (vma->vm_pgoff >= pgoff + pglen)
963 continue;
964
965 vmpglen = vma->vm_end - vma->vm_start + PAGE_SIZE - 1;
966 vmpglen >>= PAGE_SHIFT;
967 if (pgoff >= vma->vm_pgoff + vmpglen)
968 continue;
969
970 /* handle inexactly overlapping matches between mappings */
971 if (vma->vm_pgoff != pgoff || vmpglen != pglen) {
972 if (!(capabilities & BDI_CAP_MAP_DIRECT))
973 goto sharing_violation;
974 continue;
975 }
976
977 /* we've found a VMA we can share */
978 atomic_inc(&vma->vm_usage);
979
980 vml->vma = vma;
981 result = (void *) vma->vm_start;
982 goto shared;
983 }
984
David Howells165b2392007-03-22 00:11:24 -0800985 dont_share_VMAs:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 vma = NULL;
987
988 /* obtain the address at which to make a shared mapping
989 * - this is the hook for quasi-memory character devices to
990 * tell us the location of a shared mapping
991 */
992 if (file && file->f_op->get_unmapped_area) {
993 addr = file->f_op->get_unmapped_area(file, addr, len,
994 pgoff, flags);
995 if (IS_ERR((void *) addr)) {
996 ret = addr;
997 if (ret != (unsigned long) -ENOSYS)
998 goto error;
999
1000 /* the driver refused to tell us where to site
1001 * the mapping so we'll have to attempt to copy
1002 * it */
1003 ret = (unsigned long) -ENODEV;
1004 if (!(capabilities & BDI_CAP_MAP_COPY))
1005 goto error;
1006
1007 capabilities &= ~BDI_CAP_MAP_DIRECT;
1008 }
1009 }
1010 }
1011
1012 /* we're going to need a VMA struct as well */
Burman Yan4668edc2006-12-06 20:38:51 -08001013 vma = kzalloc(sizeof(struct vm_area_struct), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 if (!vma)
1015 goto error_getting_vma;
1016
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 INIT_LIST_HEAD(&vma->anon_vma_node);
1018 atomic_set(&vma->vm_usage, 1);
Matt Helsley925d1c42008-04-29 01:01:36 -07001019 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 get_file(file);
Matt Helsley925d1c42008-04-29 01:01:36 -07001021 if (vm_flags & VM_EXECUTABLE) {
1022 added_exe_file_vma(current->mm);
1023 vma->vm_mm = current->mm;
1024 }
1025 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 vma->vm_file = file;
1027 vma->vm_flags = vm_flags;
1028 vma->vm_start = addr;
1029 vma->vm_end = addr + len;
1030 vma->vm_pgoff = pgoff;
1031
1032 vml->vma = vma;
1033
1034 /* set up the mapping */
1035 if (file && vma->vm_flags & VM_SHARED)
1036 ret = do_mmap_shared_file(vma, len);
1037 else
1038 ret = do_mmap_private(vma, len);
1039 if (ret < 0)
1040 goto error;
1041
1042 /* okay... we have a mapping; now we have to register it */
1043 result = (void *) vma->vm_start;
1044
1045 if (vma->vm_flags & VM_MAPPED_COPY) {
1046 realalloc += kobjsize(result);
1047 askedalloc += len;
1048 }
1049
1050 realalloc += kobjsize(vma);
1051 askedalloc += sizeof(*vma);
1052
1053 current->mm->total_vm += len >> PAGE_SHIFT;
1054
1055 add_nommu_vma(vma);
1056
1057 shared:
1058 realalloc += kobjsize(vml);
1059 askedalloc += sizeof(*vml);
1060
David Howells30340972006-09-27 01:50:20 -07001061 add_vma_to_mm(current->mm, vml);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062
1063 up_write(&nommu_vma_sem);
1064
1065 if (prot & PROT_EXEC)
1066 flush_icache_range((unsigned long) result,
1067 (unsigned long) result + len);
1068
1069#ifdef DEBUG
1070 printk("do_mmap:\n");
1071 show_process_blocks();
1072#endif
1073
1074 return (unsigned long) result;
1075
1076 error:
1077 up_write(&nommu_vma_sem);
1078 kfree(vml);
1079 if (vma) {
Matt Helsley925d1c42008-04-29 01:01:36 -07001080 if (vma->vm_file) {
Gavin Lambert3fcd03e2006-09-30 23:27:01 -07001081 fput(vma->vm_file);
Matt Helsley925d1c42008-04-29 01:01:36 -07001082 if (vma->vm_flags & VM_EXECUTABLE)
1083 removed_exe_file_vma(vma->vm_mm);
1084 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 kfree(vma);
1086 }
1087 return ret;
1088
1089 sharing_violation:
1090 up_write(&nommu_vma_sem);
1091 printk("Attempt to share mismatched mappings\n");
1092 kfree(vml);
1093 return -EINVAL;
1094
1095 error_getting_vma:
1096 up_write(&nommu_vma_sem);
1097 kfree(vml);
Greg Ungerer66aa2b42005-09-12 11:18:10 +10001098 printk("Allocation of vma for %lu byte allocation from process %d failed\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 len, current->pid);
1100 show_free_areas();
1101 return -ENOMEM;
1102
1103 error_getting_vml:
1104 printk("Allocation of vml for %lu byte allocation from process %d failed\n",
1105 len, current->pid);
1106 show_free_areas();
1107 return -ENOMEM;
1108}
Paul Mundtb5073172007-07-21 04:37:25 -07001109EXPORT_SYMBOL(do_mmap_pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110
1111/*
1112 * handle mapping disposal for uClinux
1113 */
Matt Helsley925d1c42008-04-29 01:01:36 -07001114static void put_vma(struct mm_struct *mm, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115{
1116 if (vma) {
1117 down_write(&nommu_vma_sem);
1118
1119 if (atomic_dec_and_test(&vma->vm_usage)) {
1120 delete_nommu_vma(vma);
1121
1122 if (vma->vm_ops && vma->vm_ops->close)
1123 vma->vm_ops->close(vma);
1124
1125 /* IO memory and memory shared directly out of the pagecache from
1126 * ramfs/tmpfs mustn't be released here */
1127 if (vma->vm_flags & VM_MAPPED_COPY) {
1128 realalloc -= kobjsize((void *) vma->vm_start);
1129 askedalloc -= vma->vm_end - vma->vm_start;
1130 kfree((void *) vma->vm_start);
1131 }
1132
1133 realalloc -= kobjsize(vma);
1134 askedalloc -= sizeof(*vma);
1135
Matt Helsley925d1c42008-04-29 01:01:36 -07001136 if (vma->vm_file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137 fput(vma->vm_file);
Matt Helsley925d1c42008-04-29 01:01:36 -07001138 if (vma->vm_flags & VM_EXECUTABLE)
1139 removed_exe_file_vma(mm);
1140 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141 kfree(vma);
1142 }
1143
1144 up_write(&nommu_vma_sem);
1145 }
1146}
1147
David Howells30340972006-09-27 01:50:20 -07001148/*
1149 * release a mapping
1150 * - under NOMMU conditions the parameters must match exactly to the mapping to
1151 * be removed
1152 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153int do_munmap(struct mm_struct *mm, unsigned long addr, size_t len)
1154{
1155 struct vm_list_struct *vml, **parent;
1156 unsigned long end = addr + len;
1157
1158#ifdef DEBUG
1159 printk("do_munmap:\n");
1160#endif
1161
David Howells30340972006-09-27 01:50:20 -07001162 for (parent = &mm->context.vmlist; *parent; parent = &(*parent)->next) {
1163 if ((*parent)->vma->vm_start > addr)
1164 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165 if ((*parent)->vma->vm_start == addr &&
Greg Ungerer66aa2b42005-09-12 11:18:10 +10001166 ((len == 0) || ((*parent)->vma->vm_end == end)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167 goto found;
David Howells30340972006-09-27 01:50:20 -07001168 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169
1170 printk("munmap of non-mmaped memory by process %d (%s): %p\n",
1171 current->pid, current->comm, (void *) addr);
1172 return -EINVAL;
1173
1174 found:
1175 vml = *parent;
1176
Matt Helsley925d1c42008-04-29 01:01:36 -07001177 put_vma(mm, vml->vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178
1179 *parent = vml->next;
1180 realalloc -= kobjsize(vml);
1181 askedalloc -= sizeof(*vml);
1182 kfree(vml);
Hugh Dickins365e9c872005-10-29 18:16:18 -07001183
1184 update_hiwater_vm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185 mm->total_vm -= len >> PAGE_SHIFT;
1186
1187#ifdef DEBUG
1188 show_process_blocks();
1189#endif
1190
1191 return 0;
1192}
Paul Mundtb5073172007-07-21 04:37:25 -07001193EXPORT_SYMBOL(do_munmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194
David Howells30340972006-09-27 01:50:20 -07001195asmlinkage long sys_munmap(unsigned long addr, size_t len)
1196{
1197 int ret;
1198 struct mm_struct *mm = current->mm;
1199
1200 down_write(&mm->mmap_sem);
1201 ret = do_munmap(mm, addr, len);
1202 up_write(&mm->mmap_sem);
1203 return ret;
1204}
1205
1206/*
1207 * Release all mappings
1208 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209void exit_mmap(struct mm_struct * mm)
1210{
1211 struct vm_list_struct *tmp;
1212
1213 if (mm) {
1214#ifdef DEBUG
1215 printk("Exit_mmap:\n");
1216#endif
1217
1218 mm->total_vm = 0;
1219
1220 while ((tmp = mm->context.vmlist)) {
1221 mm->context.vmlist = tmp->next;
Matt Helsley925d1c42008-04-29 01:01:36 -07001222 put_vma(mm, tmp->vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223
1224 realalloc -= kobjsize(tmp);
1225 askedalloc -= sizeof(*tmp);
1226 kfree(tmp);
1227 }
1228
1229#ifdef DEBUG
1230 show_process_blocks();
1231#endif
1232 }
1233}
1234
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235unsigned long do_brk(unsigned long addr, unsigned long len)
1236{
1237 return -ENOMEM;
1238}
1239
1240/*
David Howells6fa5f802006-09-27 01:50:21 -07001241 * expand (or shrink) an existing mapping, potentially moving it at the same
1242 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243 *
David Howells6fa5f802006-09-27 01:50:21 -07001244 * under NOMMU conditions, we only permit changing a mapping's size, and only
1245 * as long as it stays within the hole allocated by the kmalloc() call in
1246 * do_mmap_pgoff() and the block is not shareable
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247 *
David Howells6fa5f802006-09-27 01:50:21 -07001248 * MREMAP_FIXED is not supported under NOMMU conditions
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249 */
1250unsigned long do_mremap(unsigned long addr,
1251 unsigned long old_len, unsigned long new_len,
1252 unsigned long flags, unsigned long new_addr)
1253{
David Howells6fa5f802006-09-27 01:50:21 -07001254 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255
1256 /* insanity checks first */
1257 if (new_len == 0)
1258 return (unsigned long) -EINVAL;
1259
1260 if (flags & MREMAP_FIXED && new_addr != addr)
1261 return (unsigned long) -EINVAL;
1262
David Howells6fa5f802006-09-27 01:50:21 -07001263 vma = find_vma_exact(current->mm, addr);
1264 if (!vma)
1265 return (unsigned long) -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266
David Howells6fa5f802006-09-27 01:50:21 -07001267 if (vma->vm_end != vma->vm_start + old_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 return (unsigned long) -EFAULT;
1269
David Howells6fa5f802006-09-27 01:50:21 -07001270 if (vma->vm_flags & VM_MAYSHARE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 return (unsigned long) -EPERM;
1272
1273 if (new_len > kobjsize((void *) addr))
1274 return (unsigned long) -ENOMEM;
1275
1276 /* all checks complete - do it */
David Howells6fa5f802006-09-27 01:50:21 -07001277 vma->vm_end = vma->vm_start + new_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278
1279 askedalloc -= old_len;
1280 askedalloc += new_len;
1281
David Howells6fa5f802006-09-27 01:50:21 -07001282 return vma->vm_start;
1283}
Paul Mundtb5073172007-07-21 04:37:25 -07001284EXPORT_SYMBOL(do_mremap);
David Howells6fa5f802006-09-27 01:50:21 -07001285
1286asmlinkage unsigned long sys_mremap(unsigned long addr,
1287 unsigned long old_len, unsigned long new_len,
1288 unsigned long flags, unsigned long new_addr)
1289{
1290 unsigned long ret;
1291
1292 down_write(&current->mm->mmap_sem);
1293 ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1294 up_write(&current->mm->mmap_sem);
1295 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296}
1297
Linus Torvalds6aab3412005-11-28 14:34:23 -08001298struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001299 unsigned int foll_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300{
1301 return NULL;
1302}
1303
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304int remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
1305 unsigned long to, unsigned long size, pgprot_t prot)
1306{
Greg Ungerer66aa2b42005-09-12 11:18:10 +10001307 vma->vm_start = vma->vm_pgoff << PAGE_SHIFT;
1308 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309}
Luke Yang22c4af42006-07-14 00:24:09 -07001310EXPORT_SYMBOL(remap_pfn_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311
Paul Mundtf905bc42008-02-04 22:29:59 -08001312int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1313 unsigned long pgoff)
1314{
1315 unsigned int size = vma->vm_end - vma->vm_start;
1316
1317 if (!(vma->vm_flags & VM_USERMAP))
1318 return -EINVAL;
1319
1320 vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1321 vma->vm_end = vma->vm_start + size;
1322
1323 return 0;
1324}
1325EXPORT_SYMBOL(remap_vmalloc_range);
1326
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1328{
1329}
1330
1331unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1332 unsigned long len, unsigned long pgoff, unsigned long flags)
1333{
1334 return -ENOMEM;
1335}
1336
Wolfgang Wander1363c3c2005-06-21 17:14:49 -07001337void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338{
1339}
1340
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341void unmap_mapping_range(struct address_space *mapping,
1342 loff_t const holebegin, loff_t const holelen,
1343 int even_cows)
1344{
1345}
Luke Yang22c4af42006-07-14 00:24:09 -07001346EXPORT_SYMBOL(unmap_mapping_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347
1348/*
David Howellsd56e03c2007-03-22 00:11:23 -08001349 * ask for an unmapped area at which to create a mapping on a file
1350 */
1351unsigned long get_unmapped_area(struct file *file, unsigned long addr,
1352 unsigned long len, unsigned long pgoff,
1353 unsigned long flags)
1354{
1355 unsigned long (*get_area)(struct file *, unsigned long, unsigned long,
1356 unsigned long, unsigned long);
1357
1358 get_area = current->mm->get_unmapped_area;
1359 if (file && file->f_op && file->f_op->get_unmapped_area)
1360 get_area = file->f_op->get_unmapped_area;
1361
1362 if (!get_area)
1363 return -ENOSYS;
1364
1365 return get_area(file, addr, len, pgoff, flags);
1366}
David Howellsd56e03c2007-03-22 00:11:23 -08001367EXPORT_SYMBOL(get_unmapped_area);
1368
1369/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 * Check that a process has enough memory to allocate a new virtual
1371 * mapping. 0 means there is enough memory for the allocation to
1372 * succeed and -ENOMEM implies there is not.
1373 *
1374 * We currently support three overcommit policies, which are set via the
1375 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
1376 *
1377 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1378 * Additional code 2002 Jul 20 by Robert Love.
1379 *
1380 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1381 *
1382 * Note this is a helper function intended to be used by LSMs which
1383 * wish to use this logic.
1384 */
Alan Cox34b4e4a2007-08-22 14:01:28 -07001385int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386{
1387 unsigned long free, allowed;
1388
1389 vm_acct_memory(pages);
1390
1391 /*
1392 * Sometimes we want to use more memory than we have
1393 */
1394 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1395 return 0;
1396
1397 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1398 unsigned long n;
1399
Christoph Lameter347ce432006-06-30 01:55:35 -07001400 free = global_page_state(NR_FILE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 free += nr_swap_pages;
1402
1403 /*
1404 * Any slabs which are created with the
1405 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1406 * which are reclaimable, under pressure. The dentry
1407 * cache and most inode caches should fall into this
1408 */
Christoph Lameter972d1a72006-09-25 23:31:51 -07001409 free += global_page_state(NR_SLAB_RECLAIMABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410
1411 /*
1412 * Leave the last 3% for root
1413 */
1414 if (!cap_sys_admin)
1415 free -= free / 32;
1416
1417 if (free > pages)
1418 return 0;
1419
1420 /*
1421 * nr_free_pages() is very expensive on large systems,
1422 * only call if we're about to fail.
1423 */
1424 n = nr_free_pages();
Hideo AOKId5ddc792006-04-10 22:53:01 -07001425
1426 /*
1427 * Leave reserved pages. The pages are not for anonymous pages.
1428 */
1429 if (n <= totalreserve_pages)
1430 goto error;
1431 else
1432 n -= totalreserve_pages;
1433
1434 /*
1435 * Leave the last 3% for root
1436 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437 if (!cap_sys_admin)
1438 n -= n / 32;
1439 free += n;
1440
1441 if (free > pages)
1442 return 0;
Hideo AOKId5ddc792006-04-10 22:53:01 -07001443
1444 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445 }
1446
1447 allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1448 /*
1449 * Leave the last 3% for root
1450 */
1451 if (!cap_sys_admin)
1452 allowed -= allowed / 32;
1453 allowed += total_swap_pages;
1454
1455 /* Don't let a single process grow too big:
1456 leave 3% of the size of this process for other processes */
Alan Cox731572d2008-10-29 14:01:20 -07001457 if (mm)
1458 allowed -= mm->total_vm / 32;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459
Simon Derr2f60f8d2005-08-04 19:52:03 -07001460 /*
1461 * cast `allowed' as a signed long because vm_committed_space
1462 * sometimes has a negative value
1463 */
Alan Cox80119ef2008-05-23 13:04:31 -07001464 if (atomic_long_read(&vm_committed_space) < (long)allowed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 return 0;
Hideo AOKId5ddc792006-04-10 22:53:01 -07001466error:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 vm_unacct_memory(pages);
1468
1469 return -ENOMEM;
1470}
1471
1472int in_gate_area_no_task(unsigned long addr)
1473{
1474 return 0;
1475}
David Howellsb0e15192006-01-06 00:11:42 -08001476
Nick Piggind0217ac2007-07-19 01:47:03 -07001477int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
David Howellsb0e15192006-01-06 00:11:42 -08001478{
1479 BUG();
Nick Piggind0217ac2007-07-19 01:47:03 -07001480 return 0;
David Howellsb0e15192006-01-06 00:11:42 -08001481}
Paul Mundtb5073172007-07-21 04:37:25 -07001482EXPORT_SYMBOL(filemap_fault);
David Howells0ec76a12006-09-27 01:50:15 -07001483
1484/*
1485 * Access another process' address space.
1486 * - source/target buffer must be kernel space
1487 */
1488int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
1489{
David Howells0ec76a12006-09-27 01:50:15 -07001490 struct vm_area_struct *vma;
1491 struct mm_struct *mm;
1492
1493 if (addr + len < addr)
1494 return 0;
1495
1496 mm = get_task_mm(tsk);
1497 if (!mm)
1498 return 0;
1499
1500 down_read(&mm->mmap_sem);
1501
1502 /* the access must start within one of the target process's mappings */
David Howells0159b142006-09-27 01:50:16 -07001503 vma = find_vma(mm, addr);
1504 if (vma) {
David Howells0ec76a12006-09-27 01:50:15 -07001505 /* don't overrun this mapping */
1506 if (addr + len >= vma->vm_end)
1507 len = vma->vm_end - addr;
1508
1509 /* only read or write mappings where it is permitted */
David Howellsd00c7b992006-09-27 01:50:19 -07001510 if (write && vma->vm_flags & VM_MAYWRITE)
David Howells0ec76a12006-09-27 01:50:15 -07001511 len -= copy_to_user((void *) addr, buf, len);
David Howellsd00c7b992006-09-27 01:50:19 -07001512 else if (!write && vma->vm_flags & VM_MAYREAD)
David Howells0ec76a12006-09-27 01:50:15 -07001513 len -= copy_from_user(buf, (void *) addr, len);
1514 else
1515 len = 0;
1516 } else {
1517 len = 0;
1518 }
1519
1520 up_read(&mm->mmap_sem);
1521 mmput(mm);
1522 return len;
1523}