security/ cleanups
[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(num_physpages);
48
49 /* list of shareable VMAs */
50 struct rb_root nommu_vma_tree = RB_ROOT;
51 DECLARE_RWSEM(nommu_vma_sem);
52
53 struct vm_operations_struct generic_file_vm_ops = {
54 };
55
56 /*
57  * Handle all mappings that got truncated by a "truncate()"
58  * system call.
59  *
60  * NOTE! We have to be ready to update the memory sharing
61  * between the file and the memory map for a potential last
62  * incomplete page.  Ugly, but necessary.
63  */
64 int vmtruncate(struct inode *inode, loff_t offset)
65 {
66         struct address_space *mapping = inode->i_mapping;
67         unsigned long limit;
68
69         if (inode->i_size < offset)
70                 goto do_expand;
71         i_size_write(inode, offset);
72
73         truncate_inode_pages(mapping, offset);
74         goto out_truncate;
75
76 do_expand:
77         limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
78         if (limit != RLIM_INFINITY && offset > limit)
79                 goto out_sig;
80         if (offset > inode->i_sb->s_maxbytes)
81                 goto out;
82         i_size_write(inode, offset);
83
84 out_truncate:
85         if (inode->i_op && inode->i_op->truncate)
86                 inode->i_op->truncate(inode);
87         return 0;
88 out_sig:
89         send_sig(SIGXFSZ, current, 0);
90 out:
91         return -EFBIG;
92 }
93
94 EXPORT_SYMBOL(vmtruncate);
95
96 /*
97  * Return the total memory allocated for this pointer, not
98  * just what the caller asked for.
99  *
100  * Doesn't have to be accurate, i.e. may have races.
101  */
102 unsigned int kobjsize(const void *objp)
103 {
104         struct page *page;
105
106         if (!objp || !((page = virt_to_page(objp))))
107                 return 0;
108
109         if (PageSlab(page))
110                 return ksize(objp);
111
112         BUG_ON(page->index < 0);
113         BUG_ON(page->index >= MAX_ORDER);
114
115         return (PAGE_SIZE << page->index);
116 }
117
118 /*
119  * get a list of pages in an address range belonging to the specified process
120  * and indicate the VMA that covers each page
121  * - this is potentially dodgy as we may end incrementing the page count of a
122  *   slab page or a secondary page from a compound page
123  * - don't permit access to VMAs that don't support it, such as I/O mappings
124  */
125 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
126         unsigned long start, int len, int write, int force,
127         struct page **pages, struct vm_area_struct **vmas)
128 {
129         struct vm_area_struct *vma;
130         unsigned long vm_flags;
131         int i;
132
133         /* calculate required read or write permissions.
134          * - if 'force' is set, we only require the "MAY" flags.
135          */
136         vm_flags  = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
137         vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
138
139         for (i = 0; i < len; i++) {
140                 vma = find_vma(mm, start);
141                 if (!vma)
142                         goto finish_or_fault;
143
144                 /* protect what we can, including chardevs */
145                 if (vma->vm_flags & (VM_IO | VM_PFNMAP) ||
146                     !(vm_flags & vma->vm_flags))
147                         goto finish_or_fault;
148
149                 if (pages) {
150                         pages[i] = virt_to_page(start);
151                         if (pages[i])
152                                 page_cache_get(pages[i]);
153                 }
154                 if (vmas)
155                         vmas[i] = vma;
156                 start += PAGE_SIZE;
157         }
158
159         return i;
160
161 finish_or_fault:
162         return i ? : -EFAULT;
163 }
164 EXPORT_SYMBOL(get_user_pages);
165
166 DEFINE_RWLOCK(vmlist_lock);
167 struct vm_struct *vmlist;
168
169 void vfree(void *addr)
170 {
171         kfree(addr);
172 }
173 EXPORT_SYMBOL(vfree);
174
175 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
176 {
177         /*
178          * kmalloc doesn't like __GFP_HIGHMEM for some reason
179          */
180         return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
181 }
182 EXPORT_SYMBOL(__vmalloc);
183
184 struct page * vmalloc_to_page(void *addr)
185 {
186         return virt_to_page(addr);
187 }
188 EXPORT_SYMBOL(vmalloc_to_page);
189
190 unsigned long vmalloc_to_pfn(void *addr)
191 {
192         return page_to_pfn(virt_to_page(addr));
193 }
194 EXPORT_SYMBOL(vmalloc_to_pfn);
195
196 long vread(char *buf, char *addr, unsigned long count)
197 {
198         memcpy(buf, addr, count);
199         return count;
200 }
201
202 long vwrite(char *buf, char *addr, unsigned long count)
203 {
204         /* Don't allow overflow */
205         if ((unsigned long) addr + count < count)
206                 count = -(unsigned long) addr;
207
208         memcpy(addr, buf, count);
209         return(count);
210 }
211
212 /*
213  *      vmalloc  -  allocate virtually continguos memory
214  *
215  *      @size:          allocation size
216  *
217  *      Allocate enough pages to cover @size from the page level
218  *      allocator and map them into continguos kernel virtual space.
219  *
220  *      For tight control over page level allocator and protection flags
221  *      use __vmalloc() instead.
222  */
223 void *vmalloc(unsigned long size)
224 {
225        return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
226 }
227 EXPORT_SYMBOL(vmalloc);
228
229 void *vmalloc_node(unsigned long size, int node)
230 {
231         return vmalloc(size);
232 }
233 EXPORT_SYMBOL(vmalloc_node);
234
235 /**
236  * vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
237  *      @size:          allocation size
238  *
239  *      Allocate enough 32bit PA addressable pages to cover @size from the
240  *      page level allocator and map them into continguos kernel virtual space.
241  */
242 void *vmalloc_32(unsigned long size)
243 {
244         return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
245 }
246 EXPORT_SYMBOL(vmalloc_32);
247
248 /**
249  * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
250  *      @size:          allocation size
251  *
252  * The resulting memory area is 32bit addressable and zeroed so it can be
253  * mapped to userspace without leaking data.
254  */
255 void *vmalloc_32_user(unsigned long size)
256 {
257         return __vmalloc(size, GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL);
258 }
259 EXPORT_SYMBOL(vmalloc_32_user);
260
261 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
262 {
263         BUG();
264         return NULL;
265 }
266 EXPORT_SYMBOL(vmap);
267
268 void vunmap(void *addr)
269 {
270         BUG();
271 }
272 EXPORT_SYMBOL(vunmap);
273
274 /*
275  * Implement a stub for vmalloc_sync_all() if the architecture chose not to
276  * have one.
277  */
278 void  __attribute__((weak)) vmalloc_sync_all(void)
279 {
280 }
281
282 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
283                    struct page *page)
284 {
285         return -EINVAL;
286 }
287 EXPORT_SYMBOL(vm_insert_page);
288
289 /*
290  *  sys_brk() for the most part doesn't need the global kernel
291  *  lock, except when an application is doing something nasty
292  *  like trying to un-brk an area that has already been mapped
293  *  to a regular file.  in this case, the unmapping will need
294  *  to invoke file system routines that need the global lock.
295  */
296 asmlinkage unsigned long sys_brk(unsigned long brk)
297 {
298         struct mm_struct *mm = current->mm;
299
300         if (brk < mm->start_brk || brk > mm->context.end_brk)
301                 return mm->brk;
302
303         if (mm->brk == brk)
304                 return mm->brk;
305
306         /*
307          * Always allow shrinking brk
308          */
309         if (brk <= mm->brk) {
310                 mm->brk = brk;
311                 return brk;
312         }
313
314         /*
315          * Ok, looks good - let it rip.
316          */
317         return mm->brk = brk;
318 }
319
320 #ifdef DEBUG
321 static void show_process_blocks(void)
322 {
323         struct vm_list_struct *vml;
324
325         printk("Process blocks %d:", current->pid);
326
327         for (vml = &current->mm->context.vmlist; vml; vml = vml->next) {
328                 printk(" %p: %p", vml, vml->vma);
329                 if (vml->vma)
330                         printk(" (%d @%lx #%d)",
331                                kobjsize((void *) vml->vma->vm_start),
332                                vml->vma->vm_start,
333                                atomic_read(&vml->vma->vm_usage));
334                 printk(vml->next ? " ->" : ".\n");
335         }
336 }
337 #endif /* DEBUG */
338
339 /*
340  * add a VMA into a process's mm_struct in the appropriate place in the list
341  * - should be called with mm->mmap_sem held writelocked
342  */
343 static void add_vma_to_mm(struct mm_struct *mm, struct vm_list_struct *vml)
344 {
345         struct vm_list_struct **ppv;
346
347         for (ppv = &current->mm->context.vmlist; *ppv; ppv = &(*ppv)->next)
348                 if ((*ppv)->vma->vm_start > vml->vma->vm_start)
349                         break;
350
351         vml->next = *ppv;
352         *ppv = vml;
353 }
354
355 /*
356  * look up the first VMA in which addr resides, NULL if none
357  * - should be called with mm->mmap_sem at least held readlocked
358  */
359 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
360 {
361         struct vm_list_struct *loop, *vml;
362
363         /* search the vm_start ordered list */
364         vml = NULL;
365         for (loop = mm->context.vmlist; loop; loop = loop->next) {
366                 if (loop->vma->vm_start > addr)
367                         break;
368                 vml = loop;
369         }
370
371         if (vml && vml->vma->vm_end > addr)
372                 return vml->vma;
373
374         return NULL;
375 }
376 EXPORT_SYMBOL(find_vma);
377
378 /*
379  * find a VMA
380  * - we don't extend stack VMAs under NOMMU conditions
381  */
382 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
383 {
384         return find_vma(mm, addr);
385 }
386
387 int expand_stack(struct vm_area_struct *vma, unsigned long address)
388 {
389         return -ENOMEM;
390 }
391
392 /*
393  * look up the first VMA exactly that exactly matches addr
394  * - should be called with mm->mmap_sem at least held readlocked
395  */
396 static inline struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
397                                                     unsigned long addr)
398 {
399         struct vm_list_struct *vml;
400
401         /* search the vm_start ordered list */
402         for (vml = mm->context.vmlist; vml; vml = vml->next) {
403                 if (vml->vma->vm_start == addr)
404                         return vml->vma;
405                 if (vml->vma->vm_start > addr)
406                         break;
407         }
408
409         return NULL;
410 }
411
412 /*
413  * find a VMA in the global tree
414  */
415 static inline struct vm_area_struct *find_nommu_vma(unsigned long start)
416 {
417         struct vm_area_struct *vma;
418         struct rb_node *n = nommu_vma_tree.rb_node;
419
420         while (n) {
421                 vma = rb_entry(n, struct vm_area_struct, vm_rb);
422
423                 if (start < vma->vm_start)
424                         n = n->rb_left;
425                 else if (start > vma->vm_start)
426                         n = n->rb_right;
427                 else
428                         return vma;
429         }
430
431         return NULL;
432 }
433
434 /*
435  * add a VMA in the global tree
436  */
437 static void add_nommu_vma(struct vm_area_struct *vma)
438 {
439         struct vm_area_struct *pvma;
440         struct address_space *mapping;
441         struct rb_node **p = &nommu_vma_tree.rb_node;
442         struct rb_node *parent = NULL;
443
444         /* add the VMA to the mapping */
445         if (vma->vm_file) {
446                 mapping = vma->vm_file->f_mapping;
447
448                 flush_dcache_mmap_lock(mapping);
449                 vma_prio_tree_insert(vma, &mapping->i_mmap);
450                 flush_dcache_mmap_unlock(mapping);
451         }
452
453         /* add the VMA to the master list */
454         while (*p) {
455                 parent = *p;
456                 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
457
458                 if (vma->vm_start < pvma->vm_start) {
459                         p = &(*p)->rb_left;
460                 }
461                 else if (vma->vm_start > pvma->vm_start) {
462                         p = &(*p)->rb_right;
463                 }
464                 else {
465                         /* mappings are at the same address - this can only
466                          * happen for shared-mem chardevs and shared file
467                          * mappings backed by ramfs/tmpfs */
468                         BUG_ON(!(pvma->vm_flags & VM_SHARED));
469
470                         if (vma < pvma)
471                                 p = &(*p)->rb_left;
472                         else if (vma > pvma)
473                                 p = &(*p)->rb_right;
474                         else
475                                 BUG();
476                 }
477         }
478
479         rb_link_node(&vma->vm_rb, parent, p);
480         rb_insert_color(&vma->vm_rb, &nommu_vma_tree);
481 }
482
483 /*
484  * delete a VMA from the global list
485  */
486 static void delete_nommu_vma(struct vm_area_struct *vma)
487 {
488         struct address_space *mapping;
489
490         /* remove the VMA from the mapping */
491         if (vma->vm_file) {
492                 mapping = vma->vm_file->f_mapping;
493
494                 flush_dcache_mmap_lock(mapping);
495                 vma_prio_tree_remove(vma, &mapping->i_mmap);
496                 flush_dcache_mmap_unlock(mapping);
497         }
498
499         /* remove from the master list */
500         rb_erase(&vma->vm_rb, &nommu_vma_tree);
501 }
502
503 /*
504  * determine whether a mapping should be permitted and, if so, what sort of
505  * mapping we're capable of supporting
506  */
507 static int validate_mmap_request(struct file *file,
508                                  unsigned long addr,
509                                  unsigned long len,
510                                  unsigned long prot,
511                                  unsigned long flags,
512                                  unsigned long pgoff,
513                                  unsigned long *_capabilities)
514 {
515         unsigned long capabilities;
516         unsigned long reqprot = prot;
517         int ret;
518
519         /* do the simple checks first */
520         if (flags & MAP_FIXED || addr) {
521                 printk(KERN_DEBUG
522                        "%d: Can't do fixed-address/overlay mmap of RAM\n",
523                        current->pid);
524                 return -EINVAL;
525         }
526
527         if ((flags & MAP_TYPE) != MAP_PRIVATE &&
528             (flags & MAP_TYPE) != MAP_SHARED)
529                 return -EINVAL;
530
531         if (!len)
532                 return -EINVAL;
533
534         /* Careful about overflows.. */
535         len = PAGE_ALIGN(len);
536         if (!len || len > TASK_SIZE)
537                 return -ENOMEM;
538
539         /* offset overflow? */
540         if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
541                 return -EOVERFLOW;
542
543         if (file) {
544                 /* validate file mapping requests */
545                 struct address_space *mapping;
546
547                 /* files must support mmap */
548                 if (!file->f_op || !file->f_op->mmap)
549                         return -ENODEV;
550
551                 /* work out if what we've got could possibly be shared
552                  * - we support chardevs that provide their own "memory"
553                  * - we support files/blockdevs that are memory backed
554                  */
555                 mapping = file->f_mapping;
556                 if (!mapping)
557                         mapping = file->f_path.dentry->d_inode->i_mapping;
558
559                 capabilities = 0;
560                 if (mapping && mapping->backing_dev_info)
561                         capabilities = mapping->backing_dev_info->capabilities;
562
563                 if (!capabilities) {
564                         /* no explicit capabilities set, so assume some
565                          * defaults */
566                         switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) {
567                         case S_IFREG:
568                         case S_IFBLK:
569                                 capabilities = BDI_CAP_MAP_COPY;
570                                 break;
571
572                         case S_IFCHR:
573                                 capabilities =
574                                         BDI_CAP_MAP_DIRECT |
575                                         BDI_CAP_READ_MAP |
576                                         BDI_CAP_WRITE_MAP;
577                                 break;
578
579                         default:
580                                 return -EINVAL;
581                         }
582                 }
583
584                 /* eliminate any capabilities that we can't support on this
585                  * device */
586                 if (!file->f_op->get_unmapped_area)
587                         capabilities &= ~BDI_CAP_MAP_DIRECT;
588                 if (!file->f_op->read)
589                         capabilities &= ~BDI_CAP_MAP_COPY;
590
591                 if (flags & MAP_SHARED) {
592                         /* do checks for writing, appending and locking */
593                         if ((prot & PROT_WRITE) &&
594                             !(file->f_mode & FMODE_WRITE))
595                                 return -EACCES;
596
597                         if (IS_APPEND(file->f_path.dentry->d_inode) &&
598                             (file->f_mode & FMODE_WRITE))
599                                 return -EACCES;
600
601                         if (locks_verify_locked(file->f_path.dentry->d_inode))
602                                 return -EAGAIN;
603
604                         if (!(capabilities & BDI_CAP_MAP_DIRECT))
605                                 return -ENODEV;
606
607                         if (((prot & PROT_READ)  && !(capabilities & BDI_CAP_READ_MAP))  ||
608                             ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
609                             ((prot & PROT_EXEC)  && !(capabilities & BDI_CAP_EXEC_MAP))
610                             ) {
611                                 printk("MAP_SHARED not completely supported on !MMU\n");
612                                 return -EINVAL;
613                         }
614
615                         /* we mustn't privatise shared mappings */
616                         capabilities &= ~BDI_CAP_MAP_COPY;
617                 }
618                 else {
619                         /* we're going to read the file into private memory we
620                          * allocate */
621                         if (!(capabilities & BDI_CAP_MAP_COPY))
622                                 return -ENODEV;
623
624                         /* we don't permit a private writable mapping to be
625                          * shared with the backing device */
626                         if (prot & PROT_WRITE)
627                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
628                 }
629
630                 /* handle executable mappings and implied executable
631                  * mappings */
632                 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
633                         if (prot & PROT_EXEC)
634                                 return -EPERM;
635                 }
636                 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
637                         /* handle implication of PROT_EXEC by PROT_READ */
638                         if (current->personality & READ_IMPLIES_EXEC) {
639                                 if (capabilities & BDI_CAP_EXEC_MAP)
640                                         prot |= PROT_EXEC;
641                         }
642                 }
643                 else if ((prot & PROT_READ) &&
644                          (prot & PROT_EXEC) &&
645                          !(capabilities & BDI_CAP_EXEC_MAP)
646                          ) {
647                         /* backing file is not executable, try to copy */
648                         capabilities &= ~BDI_CAP_MAP_DIRECT;
649                 }
650         }
651         else {
652                 /* anonymous mappings are always memory backed and can be
653                  * privately mapped
654                  */
655                 capabilities = BDI_CAP_MAP_COPY;
656
657                 /* handle PROT_EXEC implication by PROT_READ */
658                 if ((prot & PROT_READ) &&
659                     (current->personality & READ_IMPLIES_EXEC))
660                         prot |= PROT_EXEC;
661         }
662
663         /* allow the security API to have its say */
664         ret = security_file_mmap(file, reqprot, prot, flags, addr, 0);
665         if (ret < 0)
666                 return ret;
667
668         /* looks okay */
669         *_capabilities = capabilities;
670         return 0;
671 }
672
673 /*
674  * we've determined that we can make the mapping, now translate what we
675  * now know into VMA flags
676  */
677 static unsigned long determine_vm_flags(struct file *file,
678                                         unsigned long prot,
679                                         unsigned long flags,
680                                         unsigned long capabilities)
681 {
682         unsigned long vm_flags;
683
684         vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
685         vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
686         /* vm_flags |= mm->def_flags; */
687
688         if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
689                 /* attempt to share read-only copies of mapped file chunks */
690                 if (file && !(prot & PROT_WRITE))
691                         vm_flags |= VM_MAYSHARE;
692         }
693         else {
694                 /* overlay a shareable mapping on the backing device or inode
695                  * if possible - used for chardevs, ramfs/tmpfs/shmfs and
696                  * romfs/cramfs */
697                 if (flags & MAP_SHARED)
698                         vm_flags |= VM_MAYSHARE | VM_SHARED;
699                 else if ((((vm_flags & capabilities) ^ vm_flags) & BDI_CAP_VMFLAGS) == 0)
700                         vm_flags |= VM_MAYSHARE;
701         }
702
703         /* refuse to let anyone share private mappings with this process if
704          * it's being traced - otherwise breakpoints set in it may interfere
705          * with another untraced process
706          */
707         if ((flags & MAP_PRIVATE) && (current->ptrace & PT_PTRACED))
708                 vm_flags &= ~VM_MAYSHARE;
709
710         return vm_flags;
711 }
712
713 /*
714  * set up a shared mapping on a file
715  */
716 static int do_mmap_shared_file(struct vm_area_struct *vma, unsigned long len)
717 {
718         int ret;
719
720         ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
721         if (ret != -ENOSYS)
722                 return ret;
723
724         /* getting an ENOSYS error indicates that direct mmap isn't
725          * possible (as opposed to tried but failed) so we'll fall
726          * through to making a private copy of the data and mapping
727          * that if we can */
728         return -ENODEV;
729 }
730
731 /*
732  * set up a private mapping or an anonymous shared mapping
733  */
734 static int do_mmap_private(struct vm_area_struct *vma, unsigned long len)
735 {
736         void *base;
737         int ret;
738
739         /* invoke the file's mapping function so that it can keep track of
740          * shared mappings on devices or memory
741          * - VM_MAYSHARE will be set if it may attempt to share
742          */
743         if (vma->vm_file) {
744                 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
745                 if (ret != -ENOSYS) {
746                         /* shouldn't return success if we're not sharing */
747                         BUG_ON(ret == 0 && !(vma->vm_flags & VM_MAYSHARE));
748                         return ret; /* success or a real error */
749                 }
750
751                 /* getting an ENOSYS error indicates that direct mmap isn't
752                  * possible (as opposed to tried but failed) so we'll try to
753                  * make a private copy of the data and map that instead */
754         }
755
756         /* allocate some memory to hold the mapping
757          * - note that this may not return a page-aligned address if the object
758          *   we're allocating is smaller than a page
759          */
760         base = kmalloc(len, GFP_KERNEL|__GFP_COMP);
761         if (!base)
762                 goto enomem;
763
764         vma->vm_start = (unsigned long) base;
765         vma->vm_end = vma->vm_start + len;
766         vma->vm_flags |= VM_MAPPED_COPY;
767
768 #ifdef WARN_ON_SLACK
769         if (len + WARN_ON_SLACK <= kobjsize(result))
770                 printk("Allocation of %lu bytes from process %d has %lu bytes of slack\n",
771                        len, current->pid, kobjsize(result) - len);
772 #endif
773
774         if (vma->vm_file) {
775                 /* read the contents of a file into the copy */
776                 mm_segment_t old_fs;
777                 loff_t fpos;
778
779                 fpos = vma->vm_pgoff;
780                 fpos <<= PAGE_SHIFT;
781
782                 old_fs = get_fs();
783                 set_fs(KERNEL_DS);
784                 ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos);
785                 set_fs(old_fs);
786
787                 if (ret < 0)
788                         goto error_free;
789
790                 /* clear the last little bit */
791                 if (ret < len)
792                         memset(base + ret, 0, len - ret);
793
794         } else {
795                 /* if it's an anonymous mapping, then just clear it */
796                 memset(base, 0, len);
797         }
798
799         return 0;
800
801 error_free:
802         kfree(base);
803         vma->vm_start = 0;
804         return ret;
805
806 enomem:
807         printk("Allocation of length %lu from process %d failed\n",
808                len, current->pid);
809         show_free_areas();
810         return -ENOMEM;
811 }
812
813 /*
814  * handle mapping creation for uClinux
815  */
816 unsigned long do_mmap_pgoff(struct file *file,
817                             unsigned long addr,
818                             unsigned long len,
819                             unsigned long prot,
820                             unsigned long flags,
821                             unsigned long pgoff)
822 {
823         struct vm_list_struct *vml = NULL;
824         struct vm_area_struct *vma = NULL;
825         struct rb_node *rb;
826         unsigned long capabilities, vm_flags;
827         void *result;
828         int ret;
829
830         /* decide whether we should attempt the mapping, and if so what sort of
831          * mapping */
832         ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
833                                     &capabilities);
834         if (ret < 0)
835                 return ret;
836
837         /* we've determined that we can make the mapping, now translate what we
838          * now know into VMA flags */
839         vm_flags = determine_vm_flags(file, prot, flags, capabilities);
840
841         /* we're going to need to record the mapping if it works */
842         vml = kzalloc(sizeof(struct vm_list_struct), GFP_KERNEL);
843         if (!vml)
844                 goto error_getting_vml;
845
846         down_write(&nommu_vma_sem);
847
848         /* if we want to share, we need to check for VMAs created by other
849          * mmap() calls that overlap with our proposed mapping
850          * - we can only share with an exact match on most regular files
851          * - shared mappings on character devices and memory backed files are
852          *   permitted to overlap inexactly as far as we are concerned for in
853          *   these cases, sharing is handled in the driver or filesystem rather
854          *   than here
855          */
856         if (vm_flags & VM_MAYSHARE) {
857                 unsigned long pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
858                 unsigned long vmpglen;
859
860                 /* suppress VMA sharing for shared regions */
861                 if (vm_flags & VM_SHARED &&
862                     capabilities & BDI_CAP_MAP_DIRECT)
863                         goto dont_share_VMAs;
864
865                 for (rb = rb_first(&nommu_vma_tree); rb; rb = rb_next(rb)) {
866                         vma = rb_entry(rb, struct vm_area_struct, vm_rb);
867
868                         if (!(vma->vm_flags & VM_MAYSHARE))
869                                 continue;
870
871                         /* search for overlapping mappings on the same file */
872                         if (vma->vm_file->f_path.dentry->d_inode != file->f_path.dentry->d_inode)
873                                 continue;
874
875                         if (vma->vm_pgoff >= pgoff + pglen)
876                                 continue;
877
878                         vmpglen = vma->vm_end - vma->vm_start + PAGE_SIZE - 1;
879                         vmpglen >>= PAGE_SHIFT;
880                         if (pgoff >= vma->vm_pgoff + vmpglen)
881                                 continue;
882
883                         /* handle inexactly overlapping matches between mappings */
884                         if (vma->vm_pgoff != pgoff || vmpglen != pglen) {
885                                 if (!(capabilities & BDI_CAP_MAP_DIRECT))
886                                         goto sharing_violation;
887                                 continue;
888                         }
889
890                         /* we've found a VMA we can share */
891                         atomic_inc(&vma->vm_usage);
892
893                         vml->vma = vma;
894                         result = (void *) vma->vm_start;
895                         goto shared;
896                 }
897
898         dont_share_VMAs:
899                 vma = NULL;
900
901                 /* obtain the address at which to make a shared mapping
902                  * - this is the hook for quasi-memory character devices to
903                  *   tell us the location of a shared mapping
904                  */
905                 if (file && file->f_op->get_unmapped_area) {
906                         addr = file->f_op->get_unmapped_area(file, addr, len,
907                                                              pgoff, flags);
908                         if (IS_ERR((void *) addr)) {
909                                 ret = addr;
910                                 if (ret != (unsigned long) -ENOSYS)
911                                         goto error;
912
913                                 /* the driver refused to tell us where to site
914                                  * the mapping so we'll have to attempt to copy
915                                  * it */
916                                 ret = (unsigned long) -ENODEV;
917                                 if (!(capabilities & BDI_CAP_MAP_COPY))
918                                         goto error;
919
920                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
921                         }
922                 }
923         }
924
925         /* we're going to need a VMA struct as well */
926         vma = kzalloc(sizeof(struct vm_area_struct), GFP_KERNEL);
927         if (!vma)
928                 goto error_getting_vma;
929
930         INIT_LIST_HEAD(&vma->anon_vma_node);
931         atomic_set(&vma->vm_usage, 1);
932         if (file)
933                 get_file(file);
934         vma->vm_file    = file;
935         vma->vm_flags   = vm_flags;
936         vma->vm_start   = addr;
937         vma->vm_end     = addr + len;
938         vma->vm_pgoff   = pgoff;
939
940         vml->vma = vma;
941
942         /* set up the mapping */
943         if (file && vma->vm_flags & VM_SHARED)
944                 ret = do_mmap_shared_file(vma, len);
945         else
946                 ret = do_mmap_private(vma, len);
947         if (ret < 0)
948                 goto error;
949
950         /* okay... we have a mapping; now we have to register it */
951         result = (void *) vma->vm_start;
952
953         if (vma->vm_flags & VM_MAPPED_COPY) {
954                 realalloc += kobjsize(result);
955                 askedalloc += len;
956         }
957
958         realalloc += kobjsize(vma);
959         askedalloc += sizeof(*vma);
960
961         current->mm->total_vm += len >> PAGE_SHIFT;
962
963         add_nommu_vma(vma);
964
965  shared:
966         realalloc += kobjsize(vml);
967         askedalloc += sizeof(*vml);
968
969         add_vma_to_mm(current->mm, vml);
970
971         up_write(&nommu_vma_sem);
972
973         if (prot & PROT_EXEC)
974                 flush_icache_range((unsigned long) result,
975                                    (unsigned long) result + len);
976
977 #ifdef DEBUG
978         printk("do_mmap:\n");
979         show_process_blocks();
980 #endif
981
982         return (unsigned long) result;
983
984  error:
985         up_write(&nommu_vma_sem);
986         kfree(vml);
987         if (vma) {
988                 if (vma->vm_file)
989                         fput(vma->vm_file);
990                 kfree(vma);
991         }
992         return ret;
993
994  sharing_violation:
995         up_write(&nommu_vma_sem);
996         printk("Attempt to share mismatched mappings\n");
997         kfree(vml);
998         return -EINVAL;
999
1000  error_getting_vma:
1001         up_write(&nommu_vma_sem);
1002         kfree(vml);
1003         printk("Allocation of vma for %lu byte allocation from process %d failed\n",
1004                len, current->pid);
1005         show_free_areas();
1006         return -ENOMEM;
1007
1008  error_getting_vml:
1009         printk("Allocation of vml for %lu byte allocation from process %d failed\n",
1010                len, current->pid);
1011         show_free_areas();
1012         return -ENOMEM;
1013 }
1014 EXPORT_SYMBOL(do_mmap_pgoff);
1015
1016 /*
1017  * handle mapping disposal for uClinux
1018  */
1019 static void put_vma(struct vm_area_struct *vma)
1020 {
1021         if (vma) {
1022                 down_write(&nommu_vma_sem);
1023
1024                 if (atomic_dec_and_test(&vma->vm_usage)) {
1025                         delete_nommu_vma(vma);
1026
1027                         if (vma->vm_ops && vma->vm_ops->close)
1028                                 vma->vm_ops->close(vma);
1029
1030                         /* IO memory and memory shared directly out of the pagecache from
1031                          * ramfs/tmpfs mustn't be released here */
1032                         if (vma->vm_flags & VM_MAPPED_COPY) {
1033                                 realalloc -= kobjsize((void *) vma->vm_start);
1034                                 askedalloc -= vma->vm_end - vma->vm_start;
1035                                 kfree((void *) vma->vm_start);
1036                         }
1037
1038                         realalloc -= kobjsize(vma);
1039                         askedalloc -= sizeof(*vma);
1040
1041                         if (vma->vm_file)
1042                                 fput(vma->vm_file);
1043                         kfree(vma);
1044                 }
1045
1046                 up_write(&nommu_vma_sem);
1047         }
1048 }
1049
1050 /*
1051  * release a mapping
1052  * - under NOMMU conditions the parameters must match exactly to the mapping to
1053  *   be removed
1054  */
1055 int do_munmap(struct mm_struct *mm, unsigned long addr, size_t len)
1056 {
1057         struct vm_list_struct *vml, **parent;
1058         unsigned long end = addr + len;
1059
1060 #ifdef DEBUG
1061         printk("do_munmap:\n");
1062 #endif
1063
1064         for (parent = &mm->context.vmlist; *parent; parent = &(*parent)->next) {
1065                 if ((*parent)->vma->vm_start > addr)
1066                         break;
1067                 if ((*parent)->vma->vm_start == addr &&
1068                     ((len == 0) || ((*parent)->vma->vm_end == end)))
1069                         goto found;
1070         }
1071
1072         printk("munmap of non-mmaped memory by process %d (%s): %p\n",
1073                current->pid, current->comm, (void *) addr);
1074         return -EINVAL;
1075
1076  found:
1077         vml = *parent;
1078
1079         put_vma(vml->vma);
1080
1081         *parent = vml->next;
1082         realalloc -= kobjsize(vml);
1083         askedalloc -= sizeof(*vml);
1084         kfree(vml);
1085
1086         update_hiwater_vm(mm);
1087         mm->total_vm -= len >> PAGE_SHIFT;
1088
1089 #ifdef DEBUG
1090         show_process_blocks();
1091 #endif
1092
1093         return 0;
1094 }
1095 EXPORT_SYMBOL(do_munmap);
1096
1097 asmlinkage long sys_munmap(unsigned long addr, size_t len)
1098 {
1099         int ret;
1100         struct mm_struct *mm = current->mm;
1101
1102         down_write(&mm->mmap_sem);
1103         ret = do_munmap(mm, addr, len);
1104         up_write(&mm->mmap_sem);
1105         return ret;
1106 }
1107
1108 /*
1109  * Release all mappings
1110  */
1111 void exit_mmap(struct mm_struct * mm)
1112 {
1113         struct vm_list_struct *tmp;
1114
1115         if (mm) {
1116 #ifdef DEBUG
1117                 printk("Exit_mmap:\n");
1118 #endif
1119
1120                 mm->total_vm = 0;
1121
1122                 while ((tmp = mm->context.vmlist)) {
1123                         mm->context.vmlist = tmp->next;
1124                         put_vma(tmp->vma);
1125
1126                         realalloc -= kobjsize(tmp);
1127                         askedalloc -= sizeof(*tmp);
1128                         kfree(tmp);
1129                 }
1130
1131 #ifdef DEBUG
1132                 show_process_blocks();
1133 #endif
1134         }
1135 }
1136
1137 unsigned long do_brk(unsigned long addr, unsigned long len)
1138 {
1139         return -ENOMEM;
1140 }
1141
1142 /*
1143  * expand (or shrink) an existing mapping, potentially moving it at the same
1144  * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1145  *
1146  * under NOMMU conditions, we only permit changing a mapping's size, and only
1147  * as long as it stays within the hole allocated by the kmalloc() call in
1148  * do_mmap_pgoff() and the block is not shareable
1149  *
1150  * MREMAP_FIXED is not supported under NOMMU conditions
1151  */
1152 unsigned long do_mremap(unsigned long addr,
1153                         unsigned long old_len, unsigned long new_len,
1154                         unsigned long flags, unsigned long new_addr)
1155 {
1156         struct vm_area_struct *vma;
1157
1158         /* insanity checks first */
1159         if (new_len == 0)
1160                 return (unsigned long) -EINVAL;
1161
1162         if (flags & MREMAP_FIXED && new_addr != addr)
1163                 return (unsigned long) -EINVAL;
1164
1165         vma = find_vma_exact(current->mm, addr);
1166         if (!vma)
1167                 return (unsigned long) -EINVAL;
1168
1169         if (vma->vm_end != vma->vm_start + old_len)
1170                 return (unsigned long) -EFAULT;
1171
1172         if (vma->vm_flags & VM_MAYSHARE)
1173                 return (unsigned long) -EPERM;
1174
1175         if (new_len > kobjsize((void *) addr))
1176                 return (unsigned long) -ENOMEM;
1177
1178         /* all checks complete - do it */
1179         vma->vm_end = vma->vm_start + new_len;
1180
1181         askedalloc -= old_len;
1182         askedalloc += new_len;
1183
1184         return vma->vm_start;
1185 }
1186 EXPORT_SYMBOL(do_mremap);
1187
1188 asmlinkage unsigned long sys_mremap(unsigned long addr,
1189         unsigned long old_len, unsigned long new_len,
1190         unsigned long flags, unsigned long new_addr)
1191 {
1192         unsigned long ret;
1193
1194         down_write(&current->mm->mmap_sem);
1195         ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1196         up_write(&current->mm->mmap_sem);
1197         return ret;
1198 }
1199
1200 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1201                         unsigned int foll_flags)
1202 {
1203         return NULL;
1204 }
1205
1206 int remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
1207                 unsigned long to, unsigned long size, pgprot_t prot)
1208 {
1209         vma->vm_start = vma->vm_pgoff << PAGE_SHIFT;
1210         return 0;
1211 }
1212 EXPORT_SYMBOL(remap_pfn_range);
1213
1214 void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1215 {
1216 }
1217
1218 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1219         unsigned long len, unsigned long pgoff, unsigned long flags)
1220 {
1221         return -ENOMEM;
1222 }
1223
1224 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1225 {
1226 }
1227
1228 void unmap_mapping_range(struct address_space *mapping,
1229                          loff_t const holebegin, loff_t const holelen,
1230                          int even_cows)
1231 {
1232 }
1233 EXPORT_SYMBOL(unmap_mapping_range);
1234
1235 /*
1236  * ask for an unmapped area at which to create a mapping on a file
1237  */
1238 unsigned long get_unmapped_area(struct file *file, unsigned long addr,
1239                                 unsigned long len, unsigned long pgoff,
1240                                 unsigned long flags)
1241 {
1242         unsigned long (*get_area)(struct file *, unsigned long, unsigned long,
1243                                   unsigned long, unsigned long);
1244
1245         get_area = current->mm->get_unmapped_area;
1246         if (file && file->f_op && file->f_op->get_unmapped_area)
1247                 get_area = file->f_op->get_unmapped_area;
1248
1249         if (!get_area)
1250                 return -ENOSYS;
1251
1252         return get_area(file, addr, len, pgoff, flags);
1253 }
1254 EXPORT_SYMBOL(get_unmapped_area);
1255
1256 /*
1257  * Check that a process has enough memory to allocate a new virtual
1258  * mapping. 0 means there is enough memory for the allocation to
1259  * succeed and -ENOMEM implies there is not.
1260  *
1261  * We currently support three overcommit policies, which are set via the
1262  * vm.overcommit_memory sysctl.  See Documentation/vm/overcommit-accounting
1263  *
1264  * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1265  * Additional code 2002 Jul 20 by Robert Love.
1266  *
1267  * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1268  *
1269  * Note this is a helper function intended to be used by LSMs which
1270  * wish to use this logic.
1271  */
1272 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
1273 {
1274         unsigned long free, allowed;
1275
1276         vm_acct_memory(pages);
1277
1278         /*
1279          * Sometimes we want to use more memory than we have
1280          */
1281         if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1282                 return 0;
1283
1284         if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1285                 unsigned long n;
1286
1287                 free = global_page_state(NR_FILE_PAGES);
1288                 free += nr_swap_pages;
1289
1290                 /*
1291                  * Any slabs which are created with the
1292                  * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1293                  * which are reclaimable, under pressure.  The dentry
1294                  * cache and most inode caches should fall into this
1295                  */
1296                 free += global_page_state(NR_SLAB_RECLAIMABLE);
1297
1298                 /*
1299                  * Leave the last 3% for root
1300                  */
1301                 if (!cap_sys_admin)
1302                         free -= free / 32;
1303
1304                 if (free > pages)
1305                         return 0;
1306
1307                 /*
1308                  * nr_free_pages() is very expensive on large systems,
1309                  * only call if we're about to fail.
1310                  */
1311                 n = nr_free_pages();
1312
1313                 /*
1314                  * Leave reserved pages. The pages are not for anonymous pages.
1315                  */
1316                 if (n <= totalreserve_pages)
1317                         goto error;
1318                 else
1319                         n -= totalreserve_pages;
1320
1321                 /*
1322                  * Leave the last 3% for root
1323                  */
1324                 if (!cap_sys_admin)
1325                         n -= n / 32;
1326                 free += n;
1327
1328                 if (free > pages)
1329                         return 0;
1330
1331                 goto error;
1332         }
1333
1334         allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1335         /*
1336          * Leave the last 3% for root
1337          */
1338         if (!cap_sys_admin)
1339                 allowed -= allowed / 32;
1340         allowed += total_swap_pages;
1341
1342         /* Don't let a single process grow too big:
1343            leave 3% of the size of this process for other processes */
1344         allowed -= current->mm->total_vm / 32;
1345
1346         /*
1347          * cast `allowed' as a signed long because vm_committed_space
1348          * sometimes has a negative value
1349          */
1350         if (atomic_read(&vm_committed_space) < (long)allowed)
1351                 return 0;
1352 error:
1353         vm_unacct_memory(pages);
1354
1355         return -ENOMEM;
1356 }
1357
1358 int in_gate_area_no_task(unsigned long addr)
1359 {
1360         return 0;
1361 }
1362
1363 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1364 {
1365         BUG();
1366         return 0;
1367 }
1368 EXPORT_SYMBOL(filemap_fault);
1369
1370 /*
1371  * Access another process' address space.
1372  * - source/target buffer must be kernel space
1373  */
1374 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
1375 {
1376         struct vm_area_struct *vma;
1377         struct mm_struct *mm;
1378
1379         if (addr + len < addr)
1380                 return 0;
1381
1382         mm = get_task_mm(tsk);
1383         if (!mm)
1384                 return 0;
1385
1386         down_read(&mm->mmap_sem);
1387
1388         /* the access must start within one of the target process's mappings */
1389         vma = find_vma(mm, addr);
1390         if (vma) {
1391                 /* don't overrun this mapping */
1392                 if (addr + len >= vma->vm_end)
1393                         len = vma->vm_end - addr;
1394
1395                 /* only read or write mappings where it is permitted */
1396                 if (write && vma->vm_flags & VM_MAYWRITE)
1397                         len -= copy_to_user((void *) addr, buf, len);
1398                 else if (!write && vma->vm_flags & VM_MAYREAD)
1399                         len -= copy_from_user(buf, (void *) addr, len);
1400                 else
1401                         len = 0;
1402         } else {
1403                 len = 0;
1404         }
1405
1406         up_read(&mm->mmap_sem);
1407         mmput(mm);
1408         return len;
1409 }