security: Protection for exploiting null dereference using mmap
[linux-2.6.git] / 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 /*
371  * look up the first VMA exactly that exactly matches addr
372  * - should be called with mm->mmap_sem at least held readlocked
373  */
374 static inline struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
375                                                     unsigned long addr)
376 {
377         struct vm_list_struct *vml;
378
379         /* search the vm_start ordered list */
380         for (vml = mm->context.vmlist; vml; vml = vml->next) {
381                 if (vml->vma->vm_start == addr)
382                         return vml->vma;
383                 if (vml->vma->vm_start > addr)
384                         break;
385         }
386
387         return NULL;
388 }
389
390 /*
391  * find a VMA in the global tree
392  */
393 static inline struct vm_area_struct *find_nommu_vma(unsigned long start)
394 {
395         struct vm_area_struct *vma;
396         struct rb_node *n = nommu_vma_tree.rb_node;
397
398         while (n) {
399                 vma = rb_entry(n, struct vm_area_struct, vm_rb);
400
401                 if (start < vma->vm_start)
402                         n = n->rb_left;
403                 else if (start > vma->vm_start)
404                         n = n->rb_right;
405                 else
406                         return vma;
407         }
408
409         return NULL;
410 }
411
412 /*
413  * add a VMA in the global tree
414  */
415 static void add_nommu_vma(struct vm_area_struct *vma)
416 {
417         struct vm_area_struct *pvma;
418         struct address_space *mapping;
419         struct rb_node **p = &nommu_vma_tree.rb_node;
420         struct rb_node *parent = NULL;
421
422         /* add the VMA to the mapping */
423         if (vma->vm_file) {
424                 mapping = vma->vm_file->f_mapping;
425
426                 flush_dcache_mmap_lock(mapping);
427                 vma_prio_tree_insert(vma, &mapping->i_mmap);
428                 flush_dcache_mmap_unlock(mapping);
429         }
430
431         /* add the VMA to the master list */
432         while (*p) {
433                 parent = *p;
434                 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
435
436                 if (vma->vm_start < pvma->vm_start) {
437                         p = &(*p)->rb_left;
438                 }
439                 else if (vma->vm_start > pvma->vm_start) {
440                         p = &(*p)->rb_right;
441                 }
442                 else {
443                         /* mappings are at the same address - this can only
444                          * happen for shared-mem chardevs and shared file
445                          * mappings backed by ramfs/tmpfs */
446                         BUG_ON(!(pvma->vm_flags & VM_SHARED));
447
448                         if (vma < pvma)
449                                 p = &(*p)->rb_left;
450                         else if (vma > pvma)
451                                 p = &(*p)->rb_right;
452                         else
453                                 BUG();
454                 }
455         }
456
457         rb_link_node(&vma->vm_rb, parent, p);
458         rb_insert_color(&vma->vm_rb, &nommu_vma_tree);
459 }
460
461 /*
462  * delete a VMA from the global list
463  */
464 static void delete_nommu_vma(struct vm_area_struct *vma)
465 {
466         struct address_space *mapping;
467
468         /* remove the VMA from the mapping */
469         if (vma->vm_file) {
470                 mapping = vma->vm_file->f_mapping;
471
472                 flush_dcache_mmap_lock(mapping);
473                 vma_prio_tree_remove(vma, &mapping->i_mmap);
474                 flush_dcache_mmap_unlock(mapping);
475         }
476
477         /* remove from the master list */
478         rb_erase(&vma->vm_rb, &nommu_vma_tree);
479 }
480
481 /*
482  * determine whether a mapping should be permitted and, if so, what sort of
483  * mapping we're capable of supporting
484  */
485 static int validate_mmap_request(struct file *file,
486                                  unsigned long addr,
487                                  unsigned long len,
488                                  unsigned long prot,
489                                  unsigned long flags,
490                                  unsigned long pgoff,
491                                  unsigned long *_capabilities)
492 {
493         unsigned long capabilities;
494         unsigned long reqprot = prot;
495         int ret;
496
497         /* do the simple checks first */
498         if (flags & MAP_FIXED || addr) {
499                 printk(KERN_DEBUG
500                        "%d: Can't do fixed-address/overlay mmap of RAM\n",
501                        current->pid);
502                 return -EINVAL;
503         }
504
505         if ((flags & MAP_TYPE) != MAP_PRIVATE &&
506             (flags & MAP_TYPE) != MAP_SHARED)
507                 return -EINVAL;
508
509         if (!len)
510                 return -EINVAL;
511
512         /* Careful about overflows.. */
513         len = PAGE_ALIGN(len);
514         if (!len || len > TASK_SIZE)
515                 return -ENOMEM;
516
517         /* offset overflow? */
518         if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
519                 return -EOVERFLOW;
520
521         if (file) {
522                 /* validate file mapping requests */
523                 struct address_space *mapping;
524
525                 /* files must support mmap */
526                 if (!file->f_op || !file->f_op->mmap)
527                         return -ENODEV;
528
529                 /* work out if what we've got could possibly be shared
530                  * - we support chardevs that provide their own "memory"
531                  * - we support files/blockdevs that are memory backed
532                  */
533                 mapping = file->f_mapping;
534                 if (!mapping)
535                         mapping = file->f_path.dentry->d_inode->i_mapping;
536
537                 capabilities = 0;
538                 if (mapping && mapping->backing_dev_info)
539                         capabilities = mapping->backing_dev_info->capabilities;
540
541                 if (!capabilities) {
542                         /* no explicit capabilities set, so assume some
543                          * defaults */
544                         switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) {
545                         case S_IFREG:
546                         case S_IFBLK:
547                                 capabilities = BDI_CAP_MAP_COPY;
548                                 break;
549
550                         case S_IFCHR:
551                                 capabilities =
552                                         BDI_CAP_MAP_DIRECT |
553                                         BDI_CAP_READ_MAP |
554                                         BDI_CAP_WRITE_MAP;
555                                 break;
556
557                         default:
558                                 return -EINVAL;
559                         }
560                 }
561
562                 /* eliminate any capabilities that we can't support on this
563                  * device */
564                 if (!file->f_op->get_unmapped_area)
565                         capabilities &= ~BDI_CAP_MAP_DIRECT;
566                 if (!file->f_op->read)
567                         capabilities &= ~BDI_CAP_MAP_COPY;
568
569                 if (flags & MAP_SHARED) {
570                         /* do checks for writing, appending and locking */
571                         if ((prot & PROT_WRITE) &&
572                             !(file->f_mode & FMODE_WRITE))
573                                 return -EACCES;
574
575                         if (IS_APPEND(file->f_path.dentry->d_inode) &&
576                             (file->f_mode & FMODE_WRITE))
577                                 return -EACCES;
578
579                         if (locks_verify_locked(file->f_path.dentry->d_inode))
580                                 return -EAGAIN;
581
582                         if (!(capabilities & BDI_CAP_MAP_DIRECT))
583                                 return -ENODEV;
584
585                         if (((prot & PROT_READ)  && !(capabilities & BDI_CAP_READ_MAP))  ||
586                             ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
587                             ((prot & PROT_EXEC)  && !(capabilities & BDI_CAP_EXEC_MAP))
588                             ) {
589                                 printk("MAP_SHARED not completely supported on !MMU\n");
590                                 return -EINVAL;
591                         }
592
593                         /* we mustn't privatise shared mappings */
594                         capabilities &= ~BDI_CAP_MAP_COPY;
595                 }
596                 else {
597                         /* we're going to read the file into private memory we
598                          * allocate */
599                         if (!(capabilities & BDI_CAP_MAP_COPY))
600                                 return -ENODEV;
601
602                         /* we don't permit a private writable mapping to be
603                          * shared with the backing device */
604                         if (prot & PROT_WRITE)
605                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
606                 }
607
608                 /* handle executable mappings and implied executable
609                  * mappings */
610                 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
611                         if (prot & PROT_EXEC)
612                                 return -EPERM;
613                 }
614                 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
615                         /* handle implication of PROT_EXEC by PROT_READ */
616                         if (current->personality & READ_IMPLIES_EXEC) {
617                                 if (capabilities & BDI_CAP_EXEC_MAP)
618                                         prot |= PROT_EXEC;
619                         }
620                 }
621                 else if ((prot & PROT_READ) &&
622                          (prot & PROT_EXEC) &&
623                          !(capabilities & BDI_CAP_EXEC_MAP)
624                          ) {
625                         /* backing file is not executable, try to copy */
626                         capabilities &= ~BDI_CAP_MAP_DIRECT;
627                 }
628         }
629         else {
630                 /* anonymous mappings are always memory backed and can be
631                  * privately mapped
632                  */
633                 capabilities = BDI_CAP_MAP_COPY;
634
635                 /* handle PROT_EXEC implication by PROT_READ */
636                 if ((prot & PROT_READ) &&
637                     (current->personality & READ_IMPLIES_EXEC))
638                         prot |= PROT_EXEC;
639         }
640
641         /* allow the security API to have its say */
642         ret = security_file_mmap(file, reqprot, prot, flags, addr, 0);
643         if (ret < 0)
644                 return ret;
645
646         /* looks okay */
647         *_capabilities = capabilities;
648         return 0;
649 }
650
651 /*
652  * we've determined that we can make the mapping, now translate what we
653  * now know into VMA flags
654  */
655 static unsigned long determine_vm_flags(struct file *file,
656                                         unsigned long prot,
657                                         unsigned long flags,
658                                         unsigned long capabilities)
659 {
660         unsigned long vm_flags;
661
662         vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
663         vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
664         /* vm_flags |= mm->def_flags; */
665
666         if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
667                 /* attempt to share read-only copies of mapped file chunks */
668                 if (file && !(prot & PROT_WRITE))
669                         vm_flags |= VM_MAYSHARE;
670         }
671         else {
672                 /* overlay a shareable mapping on the backing device or inode
673                  * if possible - used for chardevs, ramfs/tmpfs/shmfs and
674                  * romfs/cramfs */
675                 if (flags & MAP_SHARED)
676                         vm_flags |= VM_MAYSHARE | VM_SHARED;
677                 else if ((((vm_flags & capabilities) ^ vm_flags) & BDI_CAP_VMFLAGS) == 0)
678                         vm_flags |= VM_MAYSHARE;
679         }
680
681         /* refuse to let anyone share private mappings with this process if
682          * it's being traced - otherwise breakpoints set in it may interfere
683          * with another untraced process
684          */
685         if ((flags & MAP_PRIVATE) && (current->ptrace & PT_PTRACED))
686                 vm_flags &= ~VM_MAYSHARE;
687
688         return vm_flags;
689 }
690
691 /*
692  * set up a shared mapping on a file
693  */
694 static int do_mmap_shared_file(struct vm_area_struct *vma, unsigned long len)
695 {
696         int ret;
697
698         ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
699         if (ret != -ENOSYS)
700                 return ret;
701
702         /* getting an ENOSYS error indicates that direct mmap isn't
703          * possible (as opposed to tried but failed) so we'll fall
704          * through to making a private copy of the data and mapping
705          * that if we can */
706         return -ENODEV;
707 }
708
709 /*
710  * set up a private mapping or an anonymous shared mapping
711  */
712 static int do_mmap_private(struct vm_area_struct *vma, unsigned long len)
713 {
714         void *base;
715         int ret;
716
717         /* invoke the file's mapping function so that it can keep track of
718          * shared mappings on devices or memory
719          * - VM_MAYSHARE will be set if it may attempt to share
720          */
721         if (vma->vm_file) {
722                 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
723                 if (ret != -ENOSYS) {
724                         /* shouldn't return success if we're not sharing */
725                         BUG_ON(ret == 0 && !(vma->vm_flags & VM_MAYSHARE));
726                         return ret; /* success or a real error */
727                 }
728
729                 /* getting an ENOSYS error indicates that direct mmap isn't
730                  * possible (as opposed to tried but failed) so we'll try to
731                  * make a private copy of the data and map that instead */
732         }
733
734         /* allocate some memory to hold the mapping
735          * - note that this may not return a page-aligned address if the object
736          *   we're allocating is smaller than a page
737          */
738         base = kmalloc(len, GFP_KERNEL|__GFP_COMP);
739         if (!base)
740                 goto enomem;
741
742         vma->vm_start = (unsigned long) base;
743         vma->vm_end = vma->vm_start + len;
744         vma->vm_flags |= VM_MAPPED_COPY;
745
746 #ifdef WARN_ON_SLACK
747         if (len + WARN_ON_SLACK <= kobjsize(result))
748                 printk("Allocation of %lu bytes from process %d has %lu bytes of slack\n",
749                        len, current->pid, kobjsize(result) - len);
750 #endif
751
752         if (vma->vm_file) {
753                 /* read the contents of a file into the copy */
754                 mm_segment_t old_fs;
755                 loff_t fpos;
756
757                 fpos = vma->vm_pgoff;
758                 fpos <<= PAGE_SHIFT;
759
760                 old_fs = get_fs();
761                 set_fs(KERNEL_DS);
762                 ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos);
763                 set_fs(old_fs);
764
765                 if (ret < 0)
766                         goto error_free;
767
768                 /* clear the last little bit */
769                 if (ret < len)
770                         memset(base + ret, 0, len - ret);
771
772         } else {
773                 /* if it's an anonymous mapping, then just clear it */
774                 memset(base, 0, len);
775         }
776
777         return 0;
778
779 error_free:
780         kfree(base);
781         vma->vm_start = 0;
782         return ret;
783
784 enomem:
785         printk("Allocation of length %lu from process %d failed\n",
786                len, current->pid);
787         show_free_areas();
788         return -ENOMEM;
789 }
790
791 /*
792  * handle mapping creation for uClinux
793  */
794 unsigned long do_mmap_pgoff(struct file *file,
795                             unsigned long addr,
796                             unsigned long len,
797                             unsigned long prot,
798                             unsigned long flags,
799                             unsigned long pgoff)
800 {
801         struct vm_list_struct *vml = NULL;
802         struct vm_area_struct *vma = NULL;
803         struct rb_node *rb;
804         unsigned long capabilities, vm_flags;
805         void *result;
806         int ret;
807
808         /* decide whether we should attempt the mapping, and if so what sort of
809          * mapping */
810         ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
811                                     &capabilities);
812         if (ret < 0)
813                 return ret;
814
815         /* we've determined that we can make the mapping, now translate what we
816          * now know into VMA flags */
817         vm_flags = determine_vm_flags(file, prot, flags, capabilities);
818
819         /* we're going to need to record the mapping if it works */
820         vml = kzalloc(sizeof(struct vm_list_struct), GFP_KERNEL);
821         if (!vml)
822                 goto error_getting_vml;
823
824         down_write(&nommu_vma_sem);
825
826         /* if we want to share, we need to check for VMAs created by other
827          * mmap() calls that overlap with our proposed mapping
828          * - we can only share with an exact match on most regular files
829          * - shared mappings on character devices and memory backed files are
830          *   permitted to overlap inexactly as far as we are concerned for in
831          *   these cases, sharing is handled in the driver or filesystem rather
832          *   than here
833          */
834         if (vm_flags & VM_MAYSHARE) {
835                 unsigned long pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
836                 unsigned long vmpglen;
837
838                 /* suppress VMA sharing for shared regions */
839                 if (vm_flags & VM_SHARED &&
840                     capabilities & BDI_CAP_MAP_DIRECT)
841                         goto dont_share_VMAs;
842
843                 for (rb = rb_first(&nommu_vma_tree); rb; rb = rb_next(rb)) {
844                         vma = rb_entry(rb, struct vm_area_struct, vm_rb);
845
846                         if (!(vma->vm_flags & VM_MAYSHARE))
847                                 continue;
848
849                         /* search for overlapping mappings on the same file */
850                         if (vma->vm_file->f_path.dentry->d_inode != file->f_path.dentry->d_inode)
851                                 continue;
852
853                         if (vma->vm_pgoff >= pgoff + pglen)
854                                 continue;
855
856                         vmpglen = vma->vm_end - vma->vm_start + PAGE_SIZE - 1;
857                         vmpglen >>= PAGE_SHIFT;
858                         if (pgoff >= vma->vm_pgoff + vmpglen)
859                                 continue;
860
861                         /* handle inexactly overlapping matches between mappings */
862                         if (vma->vm_pgoff != pgoff || vmpglen != pglen) {
863                                 if (!(capabilities & BDI_CAP_MAP_DIRECT))
864                                         goto sharing_violation;
865                                 continue;
866                         }
867
868                         /* we've found a VMA we can share */
869                         atomic_inc(&vma->vm_usage);
870
871                         vml->vma = vma;
872                         result = (void *) vma->vm_start;
873                         goto shared;
874                 }
875
876         dont_share_VMAs:
877                 vma = NULL;
878
879                 /* obtain the address at which to make a shared mapping
880                  * - this is the hook for quasi-memory character devices to
881                  *   tell us the location of a shared mapping
882                  */
883                 if (file && file->f_op->get_unmapped_area) {
884                         addr = file->f_op->get_unmapped_area(file, addr, len,
885                                                              pgoff, flags);
886                         if (IS_ERR((void *) addr)) {
887                                 ret = addr;
888                                 if (ret != (unsigned long) -ENOSYS)
889                                         goto error;
890
891                                 /* the driver refused to tell us where to site
892                                  * the mapping so we'll have to attempt to copy
893                                  * it */
894                                 ret = (unsigned long) -ENODEV;
895                                 if (!(capabilities & BDI_CAP_MAP_COPY))
896                                         goto error;
897
898                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
899                         }
900                 }
901         }
902
903         /* we're going to need a VMA struct as well */
904         vma = kzalloc(sizeof(struct vm_area_struct), GFP_KERNEL);
905         if (!vma)
906                 goto error_getting_vma;
907
908         INIT_LIST_HEAD(&vma->anon_vma_node);
909         atomic_set(&vma->vm_usage, 1);
910         if (file)
911                 get_file(file);
912         vma->vm_file    = file;
913         vma->vm_flags   = vm_flags;
914         vma->vm_start   = addr;
915         vma->vm_end     = addr + len;
916         vma->vm_pgoff   = pgoff;
917
918         vml->vma = vma;
919
920         /* set up the mapping */
921         if (file && vma->vm_flags & VM_SHARED)
922                 ret = do_mmap_shared_file(vma, len);
923         else
924                 ret = do_mmap_private(vma, len);
925         if (ret < 0)
926                 goto error;
927
928         /* okay... we have a mapping; now we have to register it */
929         result = (void *) vma->vm_start;
930
931         if (vma->vm_flags & VM_MAPPED_COPY) {
932                 realalloc += kobjsize(result);
933                 askedalloc += len;
934         }
935
936         realalloc += kobjsize(vma);
937         askedalloc += sizeof(*vma);
938
939         current->mm->total_vm += len >> PAGE_SHIFT;
940
941         add_nommu_vma(vma);
942
943  shared:
944         realalloc += kobjsize(vml);
945         askedalloc += sizeof(*vml);
946
947         add_vma_to_mm(current->mm, vml);
948
949         up_write(&nommu_vma_sem);
950
951         if (prot & PROT_EXEC)
952                 flush_icache_range((unsigned long) result,
953                                    (unsigned long) result + len);
954
955 #ifdef DEBUG
956         printk("do_mmap:\n");
957         show_process_blocks();
958 #endif
959
960         return (unsigned long) result;
961
962  error:
963         up_write(&nommu_vma_sem);
964         kfree(vml);
965         if (vma) {
966                 if (vma->vm_file)
967                         fput(vma->vm_file);
968                 kfree(vma);
969         }
970         return ret;
971
972  sharing_violation:
973         up_write(&nommu_vma_sem);
974         printk("Attempt to share mismatched mappings\n");
975         kfree(vml);
976         return -EINVAL;
977
978  error_getting_vma:
979         up_write(&nommu_vma_sem);
980         kfree(vml);
981         printk("Allocation of vma for %lu byte allocation from process %d failed\n",
982                len, current->pid);
983         show_free_areas();
984         return -ENOMEM;
985
986  error_getting_vml:
987         printk("Allocation of vml for %lu byte allocation from process %d failed\n",
988                len, current->pid);
989         show_free_areas();
990         return -ENOMEM;
991 }
992
993 /*
994  * handle mapping disposal for uClinux
995  */
996 static void put_vma(struct vm_area_struct *vma)
997 {
998         if (vma) {
999                 down_write(&nommu_vma_sem);
1000
1001                 if (atomic_dec_and_test(&vma->vm_usage)) {
1002                         delete_nommu_vma(vma);
1003
1004                         if (vma->vm_ops && vma->vm_ops->close)
1005                                 vma->vm_ops->close(vma);
1006
1007                         /* IO memory and memory shared directly out of the pagecache from
1008                          * ramfs/tmpfs mustn't be released here */
1009                         if (vma->vm_flags & VM_MAPPED_COPY) {
1010                                 realalloc -= kobjsize((void *) vma->vm_start);
1011                                 askedalloc -= vma->vm_end - vma->vm_start;
1012                                 kfree((void *) vma->vm_start);
1013                         }
1014
1015                         realalloc -= kobjsize(vma);
1016                         askedalloc -= sizeof(*vma);
1017
1018                         if (vma->vm_file)
1019                                 fput(vma->vm_file);
1020                         kfree(vma);
1021                 }
1022
1023                 up_write(&nommu_vma_sem);
1024         }
1025 }
1026
1027 /*
1028  * release a mapping
1029  * - under NOMMU conditions the parameters must match exactly to the mapping to
1030  *   be removed
1031  */
1032 int do_munmap(struct mm_struct *mm, unsigned long addr, size_t len)
1033 {
1034         struct vm_list_struct *vml, **parent;
1035         unsigned long end = addr + len;
1036
1037 #ifdef DEBUG
1038         printk("do_munmap:\n");
1039 #endif
1040
1041         for (parent = &mm->context.vmlist; *parent; parent = &(*parent)->next) {
1042                 if ((*parent)->vma->vm_start > addr)
1043                         break;
1044                 if ((*parent)->vma->vm_start == addr &&
1045                     ((len == 0) || ((*parent)->vma->vm_end == end)))
1046                         goto found;
1047         }
1048
1049         printk("munmap of non-mmaped memory by process %d (%s): %p\n",
1050                current->pid, current->comm, (void *) addr);
1051         return -EINVAL;
1052
1053  found:
1054         vml = *parent;
1055
1056         put_vma(vml->vma);
1057
1058         *parent = vml->next;
1059         realalloc -= kobjsize(vml);
1060         askedalloc -= sizeof(*vml);
1061         kfree(vml);
1062
1063         update_hiwater_vm(mm);
1064         mm->total_vm -= len >> PAGE_SHIFT;
1065
1066 #ifdef DEBUG
1067         show_process_blocks();
1068 #endif
1069
1070         return 0;
1071 }
1072
1073 asmlinkage long sys_munmap(unsigned long addr, size_t len)
1074 {
1075         int ret;
1076         struct mm_struct *mm = current->mm;
1077
1078         down_write(&mm->mmap_sem);
1079         ret = do_munmap(mm, addr, len);
1080         up_write(&mm->mmap_sem);
1081         return ret;
1082 }
1083
1084 /*
1085  * Release all mappings
1086  */
1087 void exit_mmap(struct mm_struct * mm)
1088 {
1089         struct vm_list_struct *tmp;
1090
1091         if (mm) {
1092 #ifdef DEBUG
1093                 printk("Exit_mmap:\n");
1094 #endif
1095
1096                 mm->total_vm = 0;
1097
1098                 while ((tmp = mm->context.vmlist)) {
1099                         mm->context.vmlist = tmp->next;
1100                         put_vma(tmp->vma);
1101
1102                         realalloc -= kobjsize(tmp);
1103                         askedalloc -= sizeof(*tmp);
1104                         kfree(tmp);
1105                 }
1106
1107 #ifdef DEBUG
1108                 show_process_blocks();
1109 #endif
1110         }
1111 }
1112
1113 unsigned long do_brk(unsigned long addr, unsigned long len)
1114 {
1115         return -ENOMEM;
1116 }
1117
1118 /*
1119  * expand (or shrink) an existing mapping, potentially moving it at the same
1120  * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1121  *
1122  * under NOMMU conditions, we only permit changing a mapping's size, and only
1123  * as long as it stays within the hole allocated by the kmalloc() call in
1124  * do_mmap_pgoff() and the block is not shareable
1125  *
1126  * MREMAP_FIXED is not supported under NOMMU conditions
1127  */
1128 unsigned long do_mremap(unsigned long addr,
1129                         unsigned long old_len, unsigned long new_len,
1130                         unsigned long flags, unsigned long new_addr)
1131 {
1132         struct vm_area_struct *vma;
1133
1134         /* insanity checks first */
1135         if (new_len == 0)
1136                 return (unsigned long) -EINVAL;
1137
1138         if (flags & MREMAP_FIXED && new_addr != addr)
1139                 return (unsigned long) -EINVAL;
1140
1141         vma = find_vma_exact(current->mm, addr);
1142         if (!vma)
1143                 return (unsigned long) -EINVAL;
1144
1145         if (vma->vm_end != vma->vm_start + old_len)
1146                 return (unsigned long) -EFAULT;
1147
1148         if (vma->vm_flags & VM_MAYSHARE)
1149                 return (unsigned long) -EPERM;
1150
1151         if (new_len > kobjsize((void *) addr))
1152                 return (unsigned long) -ENOMEM;
1153
1154         /* all checks complete - do it */
1155         vma->vm_end = vma->vm_start + new_len;
1156
1157         askedalloc -= old_len;
1158         askedalloc += new_len;
1159
1160         return vma->vm_start;
1161 }
1162
1163 asmlinkage unsigned long sys_mremap(unsigned long addr,
1164         unsigned long old_len, unsigned long new_len,
1165         unsigned long flags, unsigned long new_addr)
1166 {
1167         unsigned long ret;
1168
1169         down_write(&current->mm->mmap_sem);
1170         ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1171         up_write(&current->mm->mmap_sem);
1172         return ret;
1173 }
1174
1175 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1176                         unsigned int foll_flags)
1177 {
1178         return NULL;
1179 }
1180
1181 int remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
1182                 unsigned long to, unsigned long size, pgprot_t prot)
1183 {
1184         vma->vm_start = vma->vm_pgoff << PAGE_SHIFT;
1185         return 0;
1186 }
1187 EXPORT_SYMBOL(remap_pfn_range);
1188
1189 void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1190 {
1191 }
1192
1193 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1194         unsigned long len, unsigned long pgoff, unsigned long flags)
1195 {
1196         return -ENOMEM;
1197 }
1198
1199 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1200 {
1201 }
1202
1203 void unmap_mapping_range(struct address_space *mapping,
1204                          loff_t const holebegin, loff_t const holelen,
1205                          int even_cows)
1206 {
1207 }
1208 EXPORT_SYMBOL(unmap_mapping_range);
1209
1210 /*
1211  * ask for an unmapped area at which to create a mapping on a file
1212  */
1213 unsigned long get_unmapped_area(struct file *file, unsigned long addr,
1214                                 unsigned long len, unsigned long pgoff,
1215                                 unsigned long flags)
1216 {
1217         unsigned long (*get_area)(struct file *, unsigned long, unsigned long,
1218                                   unsigned long, unsigned long);
1219
1220         get_area = current->mm->get_unmapped_area;
1221         if (file && file->f_op && file->f_op->get_unmapped_area)
1222                 get_area = file->f_op->get_unmapped_area;
1223
1224         if (!get_area)
1225                 return -ENOSYS;
1226
1227         return get_area(file, addr, len, pgoff, flags);
1228 }
1229
1230 EXPORT_SYMBOL(get_unmapped_area);
1231
1232 /*
1233  * Check that a process has enough memory to allocate a new virtual
1234  * mapping. 0 means there is enough memory for the allocation to
1235  * succeed and -ENOMEM implies there is not.
1236  *
1237  * We currently support three overcommit policies, which are set via the
1238  * vm.overcommit_memory sysctl.  See Documentation/vm/overcommit-accounting
1239  *
1240  * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1241  * Additional code 2002 Jul 20 by Robert Love.
1242  *
1243  * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1244  *
1245  * Note this is a helper function intended to be used by LSMs which
1246  * wish to use this logic.
1247  */
1248 int __vm_enough_memory(long pages, int cap_sys_admin)
1249 {
1250         unsigned long free, allowed;
1251
1252         vm_acct_memory(pages);
1253
1254         /*
1255          * Sometimes we want to use more memory than we have
1256          */
1257         if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1258                 return 0;
1259
1260         if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1261                 unsigned long n;
1262
1263                 free = global_page_state(NR_FILE_PAGES);
1264                 free += nr_swap_pages;
1265
1266                 /*
1267                  * Any slabs which are created with the
1268                  * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1269                  * which are reclaimable, under pressure.  The dentry
1270                  * cache and most inode caches should fall into this
1271                  */
1272                 free += global_page_state(NR_SLAB_RECLAIMABLE);
1273
1274                 /*
1275                  * Leave the last 3% for root
1276                  */
1277                 if (!cap_sys_admin)
1278                         free -= free / 32;
1279
1280                 if (free > pages)
1281                         return 0;
1282
1283                 /*
1284                  * nr_free_pages() is very expensive on large systems,
1285                  * only call if we're about to fail.
1286                  */
1287                 n = nr_free_pages();
1288
1289                 /*
1290                  * Leave reserved pages. The pages are not for anonymous pages.
1291                  */
1292                 if (n <= totalreserve_pages)
1293                         goto error;
1294                 else
1295                         n -= totalreserve_pages;
1296
1297                 /*
1298                  * Leave the last 3% for root
1299                  */
1300                 if (!cap_sys_admin)
1301                         n -= n / 32;
1302                 free += n;
1303
1304                 if (free > pages)
1305                         return 0;
1306
1307                 goto error;
1308         }
1309
1310         allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1311         /*
1312          * Leave the last 3% for root
1313          */
1314         if (!cap_sys_admin)
1315                 allowed -= allowed / 32;
1316         allowed += total_swap_pages;
1317
1318         /* Don't let a single process grow too big:
1319            leave 3% of the size of this process for other processes */
1320         allowed -= current->mm->total_vm / 32;
1321
1322         /*
1323          * cast `allowed' as a signed long because vm_committed_space
1324          * sometimes has a negative value
1325          */
1326         if (atomic_read(&vm_committed_space) < (long)allowed)
1327                 return 0;
1328 error:
1329         vm_unacct_memory(pages);
1330
1331         return -ENOMEM;
1332 }
1333
1334 int in_gate_area_no_task(unsigned long addr)
1335 {
1336         return 0;
1337 }
1338
1339 struct page *filemap_nopage(struct vm_area_struct *area,
1340                         unsigned long address, int *type)
1341 {
1342         BUG();
1343         return NULL;
1344 }
1345
1346 /*
1347  * Access another process' address space.
1348  * - source/target buffer must be kernel space
1349  */
1350 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
1351 {
1352         struct vm_area_struct *vma;
1353         struct mm_struct *mm;
1354
1355         if (addr + len < addr)
1356                 return 0;
1357
1358         mm = get_task_mm(tsk);
1359         if (!mm)
1360                 return 0;
1361
1362         down_read(&mm->mmap_sem);
1363
1364         /* the access must start within one of the target process's mappings */
1365         vma = find_vma(mm, addr);
1366         if (vma) {
1367                 /* don't overrun this mapping */
1368                 if (addr + len >= vma->vm_end)
1369                         len = vma->vm_end - addr;
1370
1371                 /* only read or write mappings where it is permitted */
1372                 if (write && vma->vm_flags & VM_MAYWRITE)
1373                         len -= copy_to_user((void *) addr, buf, len);
1374                 else if (!write && vma->vm_flags & VM_MAYREAD)
1375                         len -= copy_from_user(buf, (void *) addr, len);
1376                 else
1377                         len = 0;
1378         } else {
1379                 len = 0;
1380         }
1381
1382         up_read(&mm->mmap_sem);
1383         mmput(mm);
1384         return len;
1385 }