Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input
[linux-2.6] / mm / nommu.c
1 /*
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>
13  */
14
15 #include <linux/mm.h>
16 #include <linux/mman.h>
17 #include <linux/swap.h>
18 #include <linux/file.h>
19 #include <linux/highmem.h>
20 #include <linux/pagemap.h>
21 #include <linux/slab.h>
22 #include <linux/vmalloc.h>
23 #include <linux/ptrace.h>
24 #include <linux/blkdev.h>
25 #include <linux/backing-dev.h>
26 #include <linux/mount.h>
27 #include <linux/personality.h>
28 #include <linux/security.h>
29 #include <linux/syscalls.h>
30
31 #include <asm/uaccess.h>
32 #include <asm/tlb.h>
33 #include <asm/tlbflush.h>
34
35 void *high_memory;
36 struct page *mem_map;
37 unsigned long max_mapnr;
38 unsigned long num_physpages;
39 unsigned long askedalloc, realalloc;
40 atomic_t vm_committed_space = ATOMIC_INIT(0);
41 int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
42 int sysctl_overcommit_ratio = 50; /* default is 50% */
43 int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
44 int heap_stack_gap = 0;
45
46 EXPORT_SYMBOL(mem_map);
47 EXPORT_SYMBOL(__vm_enough_memory);
48 EXPORT_SYMBOL(num_physpages);
49
50 /* list of shareable VMAs */
51 struct rb_root nommu_vma_tree = RB_ROOT;
52 DECLARE_RWSEM(nommu_vma_sem);
53
54 struct vm_operations_struct generic_file_vm_ops = {
55 };
56
57 EXPORT_SYMBOL(vfree);
58 EXPORT_SYMBOL(vmalloc_to_page);
59 EXPORT_SYMBOL(vmalloc_32);
60 EXPORT_SYMBOL(vmap);
61 EXPORT_SYMBOL(vunmap);
62
63 /*
64  * Handle all mappings that got truncated by a "truncate()"
65  * system call.
66  *
67  * NOTE! We have to be ready to update the memory sharing
68  * between the file and the memory map for a potential last
69  * incomplete page.  Ugly, but necessary.
70  */
71 int vmtruncate(struct inode *inode, loff_t offset)
72 {
73         struct address_space *mapping = inode->i_mapping;
74         unsigned long limit;
75
76         if (inode->i_size < offset)
77                 goto do_expand;
78         i_size_write(inode, offset);
79
80         truncate_inode_pages(mapping, offset);
81         goto out_truncate;
82
83 do_expand:
84         limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
85         if (limit != RLIM_INFINITY && offset > limit)
86                 goto out_sig;
87         if (offset > inode->i_sb->s_maxbytes)
88                 goto out;
89         i_size_write(inode, offset);
90
91 out_truncate:
92         if (inode->i_op && inode->i_op->truncate)
93                 inode->i_op->truncate(inode);
94         return 0;
95 out_sig:
96         send_sig(SIGXFSZ, current, 0);
97 out:
98         return -EFBIG;
99 }
100
101 EXPORT_SYMBOL(vmtruncate);
102
103 /*
104  * Return the total memory allocated for this pointer, not
105  * just what the caller asked for.
106  *
107  * Doesn't have to be accurate, i.e. may have races.
108  */
109 unsigned int kobjsize(const void *objp)
110 {
111         struct page *page;
112
113         if (!objp || !((page = virt_to_page(objp))))
114                 return 0;
115
116         if (PageSlab(page))
117                 return ksize(objp);
118
119         BUG_ON(page->index < 0);
120         BUG_ON(page->index >= MAX_ORDER);
121
122         return (PAGE_SIZE << page->index);
123 }
124
125 /*
126  * get a list of pages in an address range belonging to the specified process
127  * and indicate the VMA that covers each page
128  * - this is potentially dodgy as we may end incrementing the page count of a
129  *   slab page or a secondary page from a compound page
130  * - don't permit access to VMAs that don't support it, such as I/O mappings
131  */
132 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
133         unsigned long start, int len, int write, int force,
134         struct page **pages, struct vm_area_struct **vmas)
135 {
136         struct vm_area_struct *vma;
137         unsigned long vm_flags;
138         int i;
139
140         /* calculate required read or write permissions.
141          * - if 'force' is set, we only require the "MAY" flags.
142          */
143         vm_flags  = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
144         vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
145
146         for (i = 0; i < len; i++) {
147                 vma = find_vma(mm, start);
148                 if (!vma)
149                         goto finish_or_fault;
150
151                 /* protect what we can, including chardevs */
152                 if (vma->vm_flags & (VM_IO | VM_PFNMAP) ||
153                     !(vm_flags & vma->vm_flags))
154                         goto finish_or_fault;
155
156                 if (pages) {
157                         pages[i] = virt_to_page(start);
158                         if (pages[i])
159                                 page_cache_get(pages[i]);
160                 }
161                 if (vmas)
162                         vmas[i] = vma;
163                 start += PAGE_SIZE;
164         }
165
166         return i;
167
168 finish_or_fault:
169         return i ? : -EFAULT;
170 }
171
172 EXPORT_SYMBOL(get_user_pages);
173
174 DEFINE_RWLOCK(vmlist_lock);
175 struct vm_struct *vmlist;
176
177 void vfree(void *addr)
178 {
179         kfree(addr);
180 }
181
182 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
183 {
184         /*
185          * kmalloc doesn't like __GFP_HIGHMEM for some reason
186          */
187         return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
188 }
189
190 struct page * vmalloc_to_page(void *addr)
191 {
192         return virt_to_page(addr);
193 }
194
195 unsigned long vmalloc_to_pfn(void *addr)
196 {
197         return page_to_pfn(virt_to_page(addr));
198 }
199
200
201 long vread(char *buf, char *addr, unsigned long count)
202 {
203         memcpy(buf, addr, count);
204         return count;
205 }
206
207 long vwrite(char *buf, char *addr, unsigned long count)
208 {
209         /* Don't allow overflow */
210         if ((unsigned long) addr + count < count)
211                 count = -(unsigned long) addr;
212
213         memcpy(addr, buf, count);
214         return(count);
215 }
216
217 /*
218  *      vmalloc  -  allocate virtually continguos memory
219  *
220  *      @size:          allocation size
221  *
222  *      Allocate enough pages to cover @size from the page level
223  *      allocator and map them into continguos kernel virtual space.
224  *
225  *      For tight control over page level allocator and protection flags
226  *      use __vmalloc() instead.
227  */
228 void *vmalloc(unsigned long size)
229 {
230        return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
231 }
232 EXPORT_SYMBOL(vmalloc);
233
234 void *vmalloc_node(unsigned long size, int node)
235 {
236         return vmalloc(size);
237 }
238 EXPORT_SYMBOL(vmalloc_node);
239
240 /*
241  *      vmalloc_32  -  allocate virtually continguos memory (32bit addressable)
242  *
243  *      @size:          allocation size
244  *
245  *      Allocate enough 32bit PA addressable pages to cover @size from the
246  *      page level allocator and map them into continguos kernel virtual space.
247  */
248 void *vmalloc_32(unsigned long size)
249 {
250         return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
251 }
252
253 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
254 {
255         BUG();
256         return NULL;
257 }
258
259 void vunmap(void *addr)
260 {
261         BUG();
262 }
263
264 /*
265  * Implement a stub for vmalloc_sync_all() if the architecture chose not to
266  * have one.
267  */
268 void  __attribute__((weak)) vmalloc_sync_all(void)
269 {
270 }
271
272 /*
273  *  sys_brk() for the most part doesn't need the global kernel
274  *  lock, except when an application is doing something nasty
275  *  like trying to un-brk an area that has already been mapped
276  *  to a regular file.  in this case, the unmapping will need
277  *  to invoke file system routines that need the global lock.
278  */
279 asmlinkage unsigned long sys_brk(unsigned long brk)
280 {
281         struct mm_struct *mm = current->mm;
282
283         if (brk < mm->start_brk || brk > mm->context.end_brk)
284                 return mm->brk;
285
286         if (mm->brk == brk)
287                 return mm->brk;
288
289         /*
290          * Always allow shrinking brk
291          */
292         if (brk <= mm->brk) {
293                 mm->brk = brk;
294                 return brk;
295         }
296
297         /*
298          * Ok, looks good - let it rip.
299          */
300         return mm->brk = brk;
301 }
302
303 #ifdef DEBUG
304 static void show_process_blocks(void)
305 {
306         struct vm_list_struct *vml;
307
308         printk("Process blocks %d:", current->pid);
309
310         for (vml = &current->mm->context.vmlist; vml; vml = vml->next) {
311                 printk(" %p: %p", vml, vml->vma);
312                 if (vml->vma)
313                         printk(" (%d @%lx #%d)",
314                                kobjsize((void *) vml->vma->vm_start),
315                                vml->vma->vm_start,
316                                atomic_read(&vml->vma->vm_usage));
317                 printk(vml->next ? " ->" : ".\n");
318         }
319 }
320 #endif /* DEBUG */
321
322 /*
323  * add a VMA into a process's mm_struct in the appropriate place in the list
324  * - should be called with mm->mmap_sem held writelocked
325  */
326 static void add_vma_to_mm(struct mm_struct *mm, struct vm_list_struct *vml)
327 {
328         struct vm_list_struct **ppv;
329
330         for (ppv = &current->mm->context.vmlist; *ppv; ppv = &(*ppv)->next)
331                 if ((*ppv)->vma->vm_start > vml->vma->vm_start)
332                         break;
333
334         vml->next = *ppv;
335         *ppv = vml;
336 }
337
338 /*
339  * look up the first VMA in which addr resides, NULL if none
340  * - should be called with mm->mmap_sem at least held readlocked
341  */
342 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
343 {
344         struct vm_list_struct *loop, *vml;
345
346         /* search the vm_start ordered list */
347         vml = NULL;
348         for (loop = mm->context.vmlist; loop; loop = loop->next) {
349                 if (loop->vma->vm_start > addr)
350                         break;
351                 vml = loop;
352         }
353
354         if (vml && vml->vma->vm_end > addr)
355                 return vml->vma;
356
357         return NULL;
358 }
359 EXPORT_SYMBOL(find_vma);
360
361 /*
362  * find a VMA
363  * - we don't extend stack VMAs under NOMMU conditions
364  */
365 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
366 {
367         return find_vma(mm, addr);
368 }
369
370 int expand_stack(struct vm_area_struct *vma, unsigned long address)
371 {
372         return -ENOMEM;
373 }
374
375 /*
376  * look up the first VMA exactly that exactly matches addr
377  * - should be called with mm->mmap_sem at least held readlocked
378  */
379 static inline struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
380                                                     unsigned long addr)
381 {
382         struct vm_list_struct *vml;
383
384         /* search the vm_start ordered list */
385         for (vml = mm->context.vmlist; vml; vml = vml->next) {
386                 if (vml->vma->vm_start == addr)
387                         return vml->vma;
388                 if (vml->vma->vm_start > addr)
389                         break;
390         }
391
392         return NULL;
393 }
394
395 /*
396  * find a VMA in the global tree
397  */
398 static inline struct vm_area_struct *find_nommu_vma(unsigned long start)
399 {
400         struct vm_area_struct *vma;
401         struct rb_node *n = nommu_vma_tree.rb_node;
402
403         while (n) {
404                 vma = rb_entry(n, struct vm_area_struct, vm_rb);
405
406                 if (start < vma->vm_start)
407                         n = n->rb_left;
408                 else if (start > vma->vm_start)
409                         n = n->rb_right;
410                 else
411                         return vma;
412         }
413
414         return NULL;
415 }
416
417 /*
418  * add a VMA in the global tree
419  */
420 static void add_nommu_vma(struct vm_area_struct *vma)
421 {
422         struct vm_area_struct *pvma;
423         struct address_space *mapping;
424         struct rb_node **p = &nommu_vma_tree.rb_node;
425         struct rb_node *parent = NULL;
426
427         /* add the VMA to the mapping */
428         if (vma->vm_file) {
429                 mapping = vma->vm_file->f_mapping;
430
431                 flush_dcache_mmap_lock(mapping);
432                 vma_prio_tree_insert(vma, &mapping->i_mmap);
433                 flush_dcache_mmap_unlock(mapping);
434         }
435
436         /* add the VMA to the master list */
437         while (*p) {
438                 parent = *p;
439                 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
440
441                 if (vma->vm_start < pvma->vm_start) {
442                         p = &(*p)->rb_left;
443                 }
444                 else if (vma->vm_start > pvma->vm_start) {
445                         p = &(*p)->rb_right;
446                 }
447                 else {
448                         /* mappings are at the same address - this can only
449                          * happen for shared-mem chardevs and shared file
450                          * mappings backed by ramfs/tmpfs */
451                         BUG_ON(!(pvma->vm_flags & VM_SHARED));
452
453                         if (vma < pvma)
454                                 p = &(*p)->rb_left;
455                         else if (vma > pvma)
456                                 p = &(*p)->rb_right;
457                         else
458                                 BUG();
459                 }
460         }
461
462         rb_link_node(&vma->vm_rb, parent, p);
463         rb_insert_color(&vma->vm_rb, &nommu_vma_tree);
464 }
465
466 /*
467  * delete a VMA from the global list
468  */
469 static void delete_nommu_vma(struct vm_area_struct *vma)
470 {
471         struct address_space *mapping;
472
473         /* remove the VMA from the mapping */
474         if (vma->vm_file) {
475                 mapping = vma->vm_file->f_mapping;
476
477                 flush_dcache_mmap_lock(mapping);
478                 vma_prio_tree_remove(vma, &mapping->i_mmap);
479                 flush_dcache_mmap_unlock(mapping);
480         }
481
482         /* remove from the master list */
483         rb_erase(&vma->vm_rb, &nommu_vma_tree);
484 }
485
486 /*
487  * determine whether a mapping should be permitted and, if so, what sort of
488  * mapping we're capable of supporting
489  */
490 static int validate_mmap_request(struct file *file,
491                                  unsigned long addr,
492                                  unsigned long len,
493                                  unsigned long prot,
494                                  unsigned long flags,
495                                  unsigned long pgoff,
496                                  unsigned long *_capabilities)
497 {
498         unsigned long capabilities;
499         unsigned long reqprot = prot;
500         int ret;
501
502         /* do the simple checks first */
503         if (flags & MAP_FIXED || addr) {
504                 printk(KERN_DEBUG
505                        "%d: Can't do fixed-address/overlay mmap of RAM\n",
506                        current->pid);
507                 return -EINVAL;
508         }
509
510         if ((flags & MAP_TYPE) != MAP_PRIVATE &&
511             (flags & MAP_TYPE) != MAP_SHARED)
512                 return -EINVAL;
513
514         if (!len)
515                 return -EINVAL;
516
517         /* Careful about overflows.. */
518         len = PAGE_ALIGN(len);
519         if (!len || len > TASK_SIZE)
520                 return -ENOMEM;
521
522         /* offset overflow? */
523         if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
524                 return -EOVERFLOW;
525
526         if (file) {
527                 /* validate file mapping requests */
528                 struct address_space *mapping;
529
530                 /* files must support mmap */
531                 if (!file->f_op || !file->f_op->mmap)
532                         return -ENODEV;
533
534                 /* work out if what we've got could possibly be shared
535                  * - we support chardevs that provide their own "memory"
536                  * - we support files/blockdevs that are memory backed
537                  */
538                 mapping = file->f_mapping;
539                 if (!mapping)
540                         mapping = file->f_path.dentry->d_inode->i_mapping;
541
542                 capabilities = 0;
543                 if (mapping && mapping->backing_dev_info)
544                         capabilities = mapping->backing_dev_info->capabilities;
545
546                 if (!capabilities) {
547                         /* no explicit capabilities set, so assume some
548                          * defaults */
549                         switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) {
550                         case S_IFREG:
551                         case S_IFBLK:
552                                 capabilities = BDI_CAP_MAP_COPY;
553                                 break;
554
555                         case S_IFCHR:
556                                 capabilities =
557                                         BDI_CAP_MAP_DIRECT |
558                                         BDI_CAP_READ_MAP |
559                                         BDI_CAP_WRITE_MAP;
560                                 break;
561
562                         default:
563                                 return -EINVAL;
564                         }
565                 }
566
567                 /* eliminate any capabilities that we can't support on this
568                  * device */
569                 if (!file->f_op->get_unmapped_area)
570                         capabilities &= ~BDI_CAP_MAP_DIRECT;
571                 if (!file->f_op->read)
572                         capabilities &= ~BDI_CAP_MAP_COPY;
573
574                 if (flags & MAP_SHARED) {
575                         /* do checks for writing, appending and locking */
576                         if ((prot & PROT_WRITE) &&
577                             !(file->f_mode & FMODE_WRITE))
578                                 return -EACCES;
579
580                         if (IS_APPEND(file->f_path.dentry->d_inode) &&
581                             (file->f_mode & FMODE_WRITE))
582                                 return -EACCES;
583
584                         if (locks_verify_locked(file->f_path.dentry->d_inode))
585                                 return -EAGAIN;
586
587                         if (!(capabilities & BDI_CAP_MAP_DIRECT))
588                                 return -ENODEV;
589
590                         if (((prot & PROT_READ)  && !(capabilities & BDI_CAP_READ_MAP))  ||
591                             ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
592                             ((prot & PROT_EXEC)  && !(capabilities & BDI_CAP_EXEC_MAP))
593                             ) {
594                                 printk("MAP_SHARED not completely supported on !MMU\n");
595                                 return -EINVAL;
596                         }
597
598                         /* we mustn't privatise shared mappings */
599                         capabilities &= ~BDI_CAP_MAP_COPY;
600                 }
601                 else {
602                         /* we're going to read the file into private memory we
603                          * allocate */
604                         if (!(capabilities & BDI_CAP_MAP_COPY))
605                                 return -ENODEV;
606
607                         /* we don't permit a private writable mapping to be
608                          * shared with the backing device */
609                         if (prot & PROT_WRITE)
610                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
611                 }
612
613                 /* handle executable mappings and implied executable
614                  * mappings */
615                 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
616                         if (prot & PROT_EXEC)
617                                 return -EPERM;
618                 }
619                 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
620                         /* handle implication of PROT_EXEC by PROT_READ */
621                         if (current->personality & READ_IMPLIES_EXEC) {
622                                 if (capabilities & BDI_CAP_EXEC_MAP)
623                                         prot |= PROT_EXEC;
624                         }
625                 }
626                 else if ((prot & PROT_READ) &&
627                          (prot & PROT_EXEC) &&
628                          !(capabilities & BDI_CAP_EXEC_MAP)
629                          ) {
630                         /* backing file is not executable, try to copy */
631                         capabilities &= ~BDI_CAP_MAP_DIRECT;
632                 }
633         }
634         else {
635                 /* anonymous mappings are always memory backed and can be
636                  * privately mapped
637                  */
638                 capabilities = BDI_CAP_MAP_COPY;
639
640                 /* handle PROT_EXEC implication by PROT_READ */
641                 if ((prot & PROT_READ) &&
642                     (current->personality & READ_IMPLIES_EXEC))
643                         prot |= PROT_EXEC;
644         }
645
646         /* allow the security API to have its say */
647         ret = security_file_mmap(file, reqprot, prot, flags, addr, 0);
648         if (ret < 0)
649                 return ret;
650
651         /* looks okay */
652         *_capabilities = capabilities;
653         return 0;
654 }
655
656 /*
657  * we've determined that we can make the mapping, now translate what we
658  * now know into VMA flags
659  */
660 static unsigned long determine_vm_flags(struct file *file,
661                                         unsigned long prot,
662                                         unsigned long flags,
663                                         unsigned long capabilities)
664 {
665         unsigned long vm_flags;
666
667         vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
668         vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
669         /* vm_flags |= mm->def_flags; */
670
671         if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
672                 /* attempt to share read-only copies of mapped file chunks */
673                 if (file && !(prot & PROT_WRITE))
674                         vm_flags |= VM_MAYSHARE;
675         }
676         else {
677                 /* overlay a shareable mapping on the backing device or inode
678                  * if possible - used for chardevs, ramfs/tmpfs/shmfs and
679                  * romfs/cramfs */
680                 if (flags & MAP_SHARED)
681                         vm_flags |= VM_MAYSHARE | VM_SHARED;
682                 else if ((((vm_flags & capabilities) ^ vm_flags) & BDI_CAP_VMFLAGS) == 0)
683                         vm_flags |= VM_MAYSHARE;
684         }
685
686         /* refuse to let anyone share private mappings with this process if
687          * it's being traced - otherwise breakpoints set in it may interfere
688          * with another untraced process
689          */
690         if ((flags & MAP_PRIVATE) && (current->ptrace & PT_PTRACED))
691                 vm_flags &= ~VM_MAYSHARE;
692
693         return vm_flags;
694 }
695
696 /*
697  * set up a shared mapping on a file
698  */
699 static int do_mmap_shared_file(struct vm_area_struct *vma, unsigned long len)
700 {
701         int ret;
702
703         ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
704         if (ret != -ENOSYS)
705                 return ret;
706
707         /* getting an ENOSYS error indicates that direct mmap isn't
708          * possible (as opposed to tried but failed) so we'll fall
709          * through to making a private copy of the data and mapping
710          * that if we can */
711         return -ENODEV;
712 }
713
714 /*
715  * set up a private mapping or an anonymous shared mapping
716  */
717 static int do_mmap_private(struct vm_area_struct *vma, unsigned long len)
718 {
719         void *base;
720         int ret;
721
722         /* invoke the file's mapping function so that it can keep track of
723          * shared mappings on devices or memory
724          * - VM_MAYSHARE will be set if it may attempt to share
725          */
726         if (vma->vm_file) {
727                 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
728                 if (ret != -ENOSYS) {
729                         /* shouldn't return success if we're not sharing */
730                         BUG_ON(ret == 0 && !(vma->vm_flags & VM_MAYSHARE));
731                         return ret; /* success or a real error */
732                 }
733
734                 /* getting an ENOSYS error indicates that direct mmap isn't
735                  * possible (as opposed to tried but failed) so we'll try to
736                  * make a private copy of the data and map that instead */
737         }
738
739         /* allocate some memory to hold the mapping
740          * - note that this may not return a page-aligned address if the object
741          *   we're allocating is smaller than a page
742          */
743         base = kmalloc(len, GFP_KERNEL|__GFP_COMP);
744         if (!base)
745                 goto enomem;
746
747         vma->vm_start = (unsigned long) base;
748         vma->vm_end = vma->vm_start + len;
749         vma->vm_flags |= VM_MAPPED_COPY;
750
751 #ifdef WARN_ON_SLACK
752         if (len + WARN_ON_SLACK <= kobjsize(result))
753                 printk("Allocation of %lu bytes from process %d has %lu bytes of slack\n",
754                        len, current->pid, kobjsize(result) - len);
755 #endif
756
757         if (vma->vm_file) {
758                 /* read the contents of a file into the copy */
759                 mm_segment_t old_fs;
760                 loff_t fpos;
761
762                 fpos = vma->vm_pgoff;
763                 fpos <<= PAGE_SHIFT;
764
765                 old_fs = get_fs();
766                 set_fs(KERNEL_DS);
767                 ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos);
768                 set_fs(old_fs);
769
770                 if (ret < 0)
771                         goto error_free;
772
773                 /* clear the last little bit */
774                 if (ret < len)
775                         memset(base + ret, 0, len - ret);
776
777         } else {
778                 /* if it's an anonymous mapping, then just clear it */
779                 memset(base, 0, len);
780         }
781
782         return 0;
783
784 error_free:
785         kfree(base);
786         vma->vm_start = 0;
787         return ret;
788
789 enomem:
790         printk("Allocation of length %lu from process %d failed\n",
791                len, current->pid);
792         show_free_areas();
793         return -ENOMEM;
794 }
795
796 /*
797  * handle mapping creation for uClinux
798  */
799 unsigned long do_mmap_pgoff(struct file *file,
800                             unsigned long addr,
801                             unsigned long len,
802                             unsigned long prot,
803                             unsigned long flags,
804                             unsigned long pgoff)
805 {
806         struct vm_list_struct *vml = NULL;
807         struct vm_area_struct *vma = NULL;
808         struct rb_node *rb;
809         unsigned long capabilities, vm_flags;
810         void *result;
811         int ret;
812
813         /* decide whether we should attempt the mapping, and if so what sort of
814          * mapping */
815         ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
816                                     &capabilities);
817         if (ret < 0)
818                 return ret;
819
820         /* we've determined that we can make the mapping, now translate what we
821          * now know into VMA flags */
822         vm_flags = determine_vm_flags(file, prot, flags, capabilities);
823
824         /* we're going to need to record the mapping if it works */
825         vml = kzalloc(sizeof(struct vm_list_struct), GFP_KERNEL);
826         if (!vml)
827                 goto error_getting_vml;
828
829         down_write(&nommu_vma_sem);
830
831         /* if we want to share, we need to check for VMAs created by other
832          * mmap() calls that overlap with our proposed mapping
833          * - we can only share with an exact match on most regular files
834          * - shared mappings on character devices and memory backed files are
835          *   permitted to overlap inexactly as far as we are concerned for in
836          *   these cases, sharing is handled in the driver or filesystem rather
837          *   than here
838          */
839         if (vm_flags & VM_MAYSHARE) {
840                 unsigned long pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
841                 unsigned long vmpglen;
842
843                 /* suppress VMA sharing for shared regions */
844                 if (vm_flags & VM_SHARED &&
845                     capabilities & BDI_CAP_MAP_DIRECT)
846                         goto dont_share_VMAs;
847
848                 for (rb = rb_first(&nommu_vma_tree); rb; rb = rb_next(rb)) {
849                         vma = rb_entry(rb, struct vm_area_struct, vm_rb);
850
851                         if (!(vma->vm_flags & VM_MAYSHARE))
852                                 continue;
853
854                         /* search for overlapping mappings on the same file */
855                         if (vma->vm_file->f_path.dentry->d_inode != file->f_path.dentry->d_inode)
856                                 continue;
857
858                         if (vma->vm_pgoff >= pgoff + pglen)
859                                 continue;
860
861                         vmpglen = vma->vm_end - vma->vm_start + PAGE_SIZE - 1;
862                         vmpglen >>= PAGE_SHIFT;
863                         if (pgoff >= vma->vm_pgoff + vmpglen)
864                                 continue;
865
866                         /* handle inexactly overlapping matches between mappings */
867                         if (vma->vm_pgoff != pgoff || vmpglen != pglen) {
868                                 if (!(capabilities & BDI_CAP_MAP_DIRECT))
869                                         goto sharing_violation;
870                                 continue;
871                         }
872
873                         /* we've found a VMA we can share */
874                         atomic_inc(&vma->vm_usage);
875
876                         vml->vma = vma;
877                         result = (void *) vma->vm_start;
878                         goto shared;
879                 }
880
881         dont_share_VMAs:
882                 vma = NULL;
883
884                 /* obtain the address at which to make a shared mapping
885                  * - this is the hook for quasi-memory character devices to
886                  *   tell us the location of a shared mapping
887                  */
888                 if (file && file->f_op->get_unmapped_area) {
889                         addr = file->f_op->get_unmapped_area(file, addr, len,
890                                                              pgoff, flags);
891                         if (IS_ERR((void *) addr)) {
892                                 ret = addr;
893                                 if (ret != (unsigned long) -ENOSYS)
894                                         goto error;
895
896                                 /* the driver refused to tell us where to site
897                                  * the mapping so we'll have to attempt to copy
898                                  * it */
899                                 ret = (unsigned long) -ENODEV;
900                                 if (!(capabilities & BDI_CAP_MAP_COPY))
901                                         goto error;
902
903                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
904                         }
905                 }
906         }
907
908         /* we're going to need a VMA struct as well */
909         vma = kzalloc(sizeof(struct vm_area_struct), GFP_KERNEL);
910         if (!vma)
911                 goto error_getting_vma;
912
913         INIT_LIST_HEAD(&vma->anon_vma_node);
914         atomic_set(&vma->vm_usage, 1);
915         if (file)
916                 get_file(file);
917         vma->vm_file    = file;
918         vma->vm_flags   = vm_flags;
919         vma->vm_start   = addr;
920         vma->vm_end     = addr + len;
921         vma->vm_pgoff   = pgoff;
922
923         vml->vma = vma;
924
925         /* set up the mapping */
926         if (file && vma->vm_flags & VM_SHARED)
927                 ret = do_mmap_shared_file(vma, len);
928         else
929                 ret = do_mmap_private(vma, len);
930         if (ret < 0)
931                 goto error;
932
933         /* okay... we have a mapping; now we have to register it */
934         result = (void *) vma->vm_start;
935
936         if (vma->vm_flags & VM_MAPPED_COPY) {
937                 realalloc += kobjsize(result);
938                 askedalloc += len;
939         }
940
941         realalloc += kobjsize(vma);
942         askedalloc += sizeof(*vma);
943
944         current->mm->total_vm += len >> PAGE_SHIFT;
945
946         add_nommu_vma(vma);
947
948  shared:
949         realalloc += kobjsize(vml);
950         askedalloc += sizeof(*vml);
951
952         add_vma_to_mm(current->mm, vml);
953
954         up_write(&nommu_vma_sem);
955
956         if (prot & PROT_EXEC)
957                 flush_icache_range((unsigned long) result,
958                                    (unsigned long) result + len);
959
960 #ifdef DEBUG
961         printk("do_mmap:\n");
962         show_process_blocks();
963 #endif
964
965         return (unsigned long) result;
966
967  error:
968         up_write(&nommu_vma_sem);
969         kfree(vml);
970         if (vma) {
971                 if (vma->vm_file)
972                         fput(vma->vm_file);
973                 kfree(vma);
974         }
975         return ret;
976
977  sharing_violation:
978         up_write(&nommu_vma_sem);
979         printk("Attempt to share mismatched mappings\n");
980         kfree(vml);
981         return -EINVAL;
982
983  error_getting_vma:
984         up_write(&nommu_vma_sem);
985         kfree(vml);
986         printk("Allocation of vma for %lu byte allocation from process %d failed\n",
987                len, current->pid);
988         show_free_areas();
989         return -ENOMEM;
990
991  error_getting_vml:
992         printk("Allocation of vml for %lu byte allocation from process %d failed\n",
993                len, current->pid);
994         show_free_areas();
995         return -ENOMEM;
996 }
997
998 /*
999  * handle mapping disposal for uClinux
1000  */
1001 static void put_vma(struct vm_area_struct *vma)
1002 {
1003         if (vma) {
1004                 down_write(&nommu_vma_sem);
1005
1006                 if (atomic_dec_and_test(&vma->vm_usage)) {
1007                         delete_nommu_vma(vma);
1008
1009                         if (vma->vm_ops && vma->vm_ops->close)
1010                                 vma->vm_ops->close(vma);
1011
1012                         /* IO memory and memory shared directly out of the pagecache from
1013                          * ramfs/tmpfs mustn't be released here */
1014                         if (vma->vm_flags & VM_MAPPED_COPY) {
1015                                 realalloc -= kobjsize((void *) vma->vm_start);
1016                                 askedalloc -= vma->vm_end - vma->vm_start;
1017                                 kfree((void *) vma->vm_start);
1018                         }
1019
1020                         realalloc -= kobjsize(vma);
1021                         askedalloc -= sizeof(*vma);
1022
1023                         if (vma->vm_file)
1024                                 fput(vma->vm_file);
1025                         kfree(vma);
1026                 }
1027
1028                 up_write(&nommu_vma_sem);
1029         }
1030 }
1031
1032 /*
1033  * release a mapping
1034  * - under NOMMU conditions the parameters must match exactly to the mapping to
1035  *   be removed
1036  */
1037 int do_munmap(struct mm_struct *mm, unsigned long addr, size_t len)
1038 {
1039         struct vm_list_struct *vml, **parent;
1040         unsigned long end = addr + len;
1041
1042 #ifdef DEBUG
1043         printk("do_munmap:\n");
1044 #endif
1045
1046         for (parent = &mm->context.vmlist; *parent; parent = &(*parent)->next) {
1047                 if ((*parent)->vma->vm_start > addr)
1048                         break;
1049                 if ((*parent)->vma->vm_start == addr &&
1050                     ((len == 0) || ((*parent)->vma->vm_end == end)))
1051                         goto found;
1052         }
1053
1054         printk("munmap of non-mmaped memory by process %d (%s): %p\n",
1055                current->pid, current->comm, (void *) addr);
1056         return -EINVAL;
1057
1058  found:
1059         vml = *parent;
1060
1061         put_vma(vml->vma);
1062
1063         *parent = vml->next;
1064         realalloc -= kobjsize(vml);
1065         askedalloc -= sizeof(*vml);
1066         kfree(vml);
1067
1068         update_hiwater_vm(mm);
1069         mm->total_vm -= len >> PAGE_SHIFT;
1070
1071 #ifdef DEBUG
1072         show_process_blocks();
1073 #endif
1074
1075         return 0;
1076 }
1077
1078 asmlinkage long sys_munmap(unsigned long addr, size_t len)
1079 {
1080         int ret;
1081         struct mm_struct *mm = current->mm;
1082
1083         down_write(&mm->mmap_sem);
1084         ret = do_munmap(mm, addr, len);
1085         up_write(&mm->mmap_sem);
1086         return ret;
1087 }
1088
1089 /*
1090  * Release all mappings
1091  */
1092 void exit_mmap(struct mm_struct * mm)
1093 {
1094         struct vm_list_struct *tmp;
1095
1096         if (mm) {
1097 #ifdef DEBUG
1098                 printk("Exit_mmap:\n");
1099 #endif
1100
1101                 mm->total_vm = 0;
1102
1103                 while ((tmp = mm->context.vmlist)) {
1104                         mm->context.vmlist = tmp->next;
1105                         put_vma(tmp->vma);
1106
1107                         realalloc -= kobjsize(tmp);
1108                         askedalloc -= sizeof(*tmp);
1109                         kfree(tmp);
1110                 }
1111
1112 #ifdef DEBUG
1113                 show_process_blocks();
1114 #endif
1115         }
1116 }
1117
1118 unsigned long do_brk(unsigned long addr, unsigned long len)
1119 {
1120         return -ENOMEM;
1121 }
1122
1123 /*
1124  * expand (or shrink) an existing mapping, potentially moving it at the same
1125  * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1126  *
1127  * under NOMMU conditions, we only permit changing a mapping's size, and only
1128  * as long as it stays within the hole allocated by the kmalloc() call in
1129  * do_mmap_pgoff() and the block is not shareable
1130  *
1131  * MREMAP_FIXED is not supported under NOMMU conditions
1132  */
1133 unsigned long do_mremap(unsigned long addr,
1134                         unsigned long old_len, unsigned long new_len,
1135                         unsigned long flags, unsigned long new_addr)
1136 {
1137         struct vm_area_struct *vma;
1138
1139         /* insanity checks first */
1140         if (new_len == 0)
1141                 return (unsigned long) -EINVAL;
1142
1143         if (flags & MREMAP_FIXED && new_addr != addr)
1144                 return (unsigned long) -EINVAL;
1145
1146         vma = find_vma_exact(current->mm, addr);
1147         if (!vma)
1148                 return (unsigned long) -EINVAL;
1149
1150         if (vma->vm_end != vma->vm_start + old_len)
1151                 return (unsigned long) -EFAULT;
1152
1153         if (vma->vm_flags & VM_MAYSHARE)
1154                 return (unsigned long) -EPERM;
1155
1156         if (new_len > kobjsize((void *) addr))
1157                 return (unsigned long) -ENOMEM;
1158
1159         /* all checks complete - do it */
1160         vma->vm_end = vma->vm_start + new_len;
1161
1162         askedalloc -= old_len;
1163         askedalloc += new_len;
1164
1165         return vma->vm_start;
1166 }
1167
1168 asmlinkage unsigned long sys_mremap(unsigned long addr,
1169         unsigned long old_len, unsigned long new_len,
1170         unsigned long flags, unsigned long new_addr)
1171 {
1172         unsigned long ret;
1173
1174         down_write(&current->mm->mmap_sem);
1175         ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1176         up_write(&current->mm->mmap_sem);
1177         return ret;
1178 }
1179
1180 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1181                         unsigned int foll_flags)
1182 {
1183         return NULL;
1184 }
1185
1186 int remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
1187                 unsigned long to, unsigned long size, pgprot_t prot)
1188 {
1189         vma->vm_start = vma->vm_pgoff << PAGE_SHIFT;
1190         return 0;
1191 }
1192 EXPORT_SYMBOL(remap_pfn_range);
1193
1194 void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1195 {
1196 }
1197
1198 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1199         unsigned long len, unsigned long pgoff, unsigned long flags)
1200 {
1201         return -ENOMEM;
1202 }
1203
1204 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1205 {
1206 }
1207
1208 void unmap_mapping_range(struct address_space *mapping,
1209                          loff_t const holebegin, loff_t const holelen,
1210                          int even_cows)
1211 {
1212 }
1213 EXPORT_SYMBOL(unmap_mapping_range);
1214
1215 /*
1216  * ask for an unmapped area at which to create a mapping on a file
1217  */
1218 unsigned long get_unmapped_area(struct file *file, unsigned long addr,
1219                                 unsigned long len, unsigned long pgoff,
1220                                 unsigned long flags)
1221 {
1222         unsigned long (*get_area)(struct file *, unsigned long, unsigned long,
1223                                   unsigned long, unsigned long);
1224
1225         get_area = current->mm->get_unmapped_area;
1226         if (file && file->f_op && file->f_op->get_unmapped_area)
1227                 get_area = file->f_op->get_unmapped_area;
1228
1229         if (!get_area)
1230                 return -ENOSYS;
1231
1232         return get_area(file, addr, len, pgoff, flags);
1233 }
1234
1235 EXPORT_SYMBOL(get_unmapped_area);
1236
1237 /*
1238  * Check that a process has enough memory to allocate a new virtual
1239  * mapping. 0 means there is enough memory for the allocation to
1240  * succeed and -ENOMEM implies there is not.
1241  *
1242  * We currently support three overcommit policies, which are set via the
1243  * vm.overcommit_memory sysctl.  See Documentation/vm/overcommit-accounting
1244  *
1245  * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1246  * Additional code 2002 Jul 20 by Robert Love.
1247  *
1248  * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1249  *
1250  * Note this is a helper function intended to be used by LSMs which
1251  * wish to use this logic.
1252  */
1253 int __vm_enough_memory(long pages, int cap_sys_admin)
1254 {
1255         unsigned long free, allowed;
1256
1257         vm_acct_memory(pages);
1258
1259         /*
1260          * Sometimes we want to use more memory than we have
1261          */
1262         if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1263                 return 0;
1264
1265         if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1266                 unsigned long n;
1267
1268                 free = global_page_state(NR_FILE_PAGES);
1269                 free += nr_swap_pages;
1270
1271                 /*
1272                  * Any slabs which are created with the
1273                  * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1274                  * which are reclaimable, under pressure.  The dentry
1275                  * cache and most inode caches should fall into this
1276                  */
1277                 free += global_page_state(NR_SLAB_RECLAIMABLE);
1278
1279                 /*
1280                  * Leave the last 3% for root
1281                  */
1282                 if (!cap_sys_admin)
1283                         free -= free / 32;
1284
1285                 if (free > pages)
1286                         return 0;
1287
1288                 /*
1289                  * nr_free_pages() is very expensive on large systems,
1290                  * only call if we're about to fail.
1291                  */
1292                 n = nr_free_pages();
1293
1294                 /*
1295                  * Leave reserved pages. The pages are not for anonymous pages.
1296                  */
1297                 if (n <= totalreserve_pages)
1298                         goto error;
1299                 else
1300                         n -= totalreserve_pages;
1301
1302                 /*
1303                  * Leave the last 3% for root
1304                  */
1305                 if (!cap_sys_admin)
1306                         n -= n / 32;
1307                 free += n;
1308
1309                 if (free > pages)
1310                         return 0;
1311
1312                 goto error;
1313         }
1314
1315         allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1316         /*
1317          * Leave the last 3% for root
1318          */
1319         if (!cap_sys_admin)
1320                 allowed -= allowed / 32;
1321         allowed += total_swap_pages;
1322
1323         /* Don't let a single process grow too big:
1324            leave 3% of the size of this process for other processes */
1325         allowed -= current->mm->total_vm / 32;
1326
1327         /*
1328          * cast `allowed' as a signed long because vm_committed_space
1329          * sometimes has a negative value
1330          */
1331         if (atomic_read(&vm_committed_space) < (long)allowed)
1332                 return 0;
1333 error:
1334         vm_unacct_memory(pages);
1335
1336         return -ENOMEM;
1337 }
1338
1339 int in_gate_area_no_task(unsigned long addr)
1340 {
1341         return 0;
1342 }
1343
1344 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1345 {
1346         BUG();
1347         return 0;
1348 }
1349
1350 /*
1351  * Access another process' address space.
1352  * - source/target buffer must be kernel space
1353  */
1354 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
1355 {
1356         struct vm_area_struct *vma;
1357         struct mm_struct *mm;
1358
1359         if (addr + len < addr)
1360                 return 0;
1361
1362         mm = get_task_mm(tsk);
1363         if (!mm)
1364                 return 0;
1365
1366         down_read(&mm->mmap_sem);
1367
1368         /* the access must start within one of the target process's mappings */
1369         vma = find_vma(mm, addr);
1370         if (vma) {
1371                 /* don't overrun this mapping */
1372                 if (addr + len >= vma->vm_end)
1373                         len = vma->vm_end - addr;
1374
1375                 /* only read or write mappings where it is permitted */
1376                 if (write && vma->vm_flags & VM_MAYWRITE)
1377                         len -= copy_to_user((void *) addr, buf, len);
1378                 else if (!write && vma->vm_flags & VM_MAYREAD)
1379                         len -= copy_from_user(buf, (void *) addr, len);
1380                 else
1381                         len = 0;
1382         } else {
1383                 len = 0;
1384         }
1385
1386         up_read(&mm->mmap_sem);
1387         mmput(mm);
1388         return len;
1389 }