nommu: add new vmalloc_user() and remap_vmalloc_range() interfaces.
[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  *  Copyright (c) 2007      Paul Mundt <lethal@linux-sh.org>
14  */
15
16 #include <linux/module.h>
17 #include <linux/mm.h>
18 #include <linux/mman.h>
19 #include <linux/swap.h>
20 #include <linux/file.h>
21 #include <linux/highmem.h>
22 #include <linux/pagemap.h>
23 #include <linux/slab.h>
24 #include <linux/vmalloc.h>
25 #include <linux/ptrace.h>
26 #include <linux/blkdev.h>
27 #include <linux/backing-dev.h>
28 #include <linux/mount.h>
29 #include <linux/personality.h>
30 #include <linux/security.h>
31 #include <linux/syscalls.h>
32
33 #include <asm/uaccess.h>
34 #include <asm/tlb.h>
35 #include <asm/tlbflush.h>
36
37 void *high_memory;
38 struct page *mem_map;
39 unsigned long max_mapnr;
40 unsigned long num_physpages;
41 unsigned long askedalloc, realalloc;
42 atomic_t vm_committed_space = ATOMIC_INIT(0);
43 int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
44 int sysctl_overcommit_ratio = 50; /* default is 50% */
45 int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
46 int heap_stack_gap = 0;
47
48 EXPORT_SYMBOL(mem_map);
49 EXPORT_SYMBOL(num_physpages);
50
51 /* list of shareable VMAs */
52 struct rb_root nommu_vma_tree = RB_ROOT;
53 DECLARE_RWSEM(nommu_vma_sem);
54
55 struct vm_operations_struct generic_file_vm_ops = {
56 };
57
58 /*
59  * Handle all mappings that got truncated by a "truncate()"
60  * system call.
61  *
62  * NOTE! We have to be ready to update the memory sharing
63  * between the file and the memory map for a potential last
64  * incomplete page.  Ugly, but necessary.
65  */
66 int vmtruncate(struct inode *inode, loff_t offset)
67 {
68         struct address_space *mapping = inode->i_mapping;
69         unsigned long limit;
70
71         if (inode->i_size < offset)
72                 goto do_expand;
73         i_size_write(inode, offset);
74
75         truncate_inode_pages(mapping, offset);
76         goto out_truncate;
77
78 do_expand:
79         limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
80         if (limit != RLIM_INFINITY && offset > limit)
81                 goto out_sig;
82         if (offset > inode->i_sb->s_maxbytes)
83                 goto out;
84         i_size_write(inode, offset);
85
86 out_truncate:
87         if (inode->i_op && inode->i_op->truncate)
88                 inode->i_op->truncate(inode);
89         return 0;
90 out_sig:
91         send_sig(SIGXFSZ, current, 0);
92 out:
93         return -EFBIG;
94 }
95
96 EXPORT_SYMBOL(vmtruncate);
97
98 /*
99  * Return the total memory allocated for this pointer, not
100  * just what the caller asked for.
101  *
102  * Doesn't have to be accurate, i.e. may have races.
103  */
104 unsigned int kobjsize(const void *objp)
105 {
106         struct page *page;
107
108         if (!objp || !((page = virt_to_page(objp))))
109                 return 0;
110
111         if (PageSlab(page))
112                 return ksize(objp);
113
114         BUG_ON(page->index < 0);
115         BUG_ON(page->index >= MAX_ORDER);
116
117         return (PAGE_SIZE << page->index);
118 }
119
120 /*
121  * get a list of pages in an address range belonging to the specified process
122  * and indicate the VMA that covers each page
123  * - this is potentially dodgy as we may end incrementing the page count of a
124  *   slab page or a secondary page from a compound page
125  * - don't permit access to VMAs that don't support it, such as I/O mappings
126  */
127 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
128         unsigned long start, int len, int write, int force,
129         struct page **pages, struct vm_area_struct **vmas)
130 {
131         struct vm_area_struct *vma;
132         unsigned long vm_flags;
133         int i;
134
135         /* calculate required read or write permissions.
136          * - if 'force' is set, we only require the "MAY" flags.
137          */
138         vm_flags  = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
139         vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
140
141         for (i = 0; i < len; i++) {
142                 vma = find_vma(mm, start);
143                 if (!vma)
144                         goto finish_or_fault;
145
146                 /* protect what we can, including chardevs */
147                 if (vma->vm_flags & (VM_IO | VM_PFNMAP) ||
148                     !(vm_flags & vma->vm_flags))
149                         goto finish_or_fault;
150
151                 if (pages) {
152                         pages[i] = virt_to_page(start);
153                         if (pages[i])
154                                 page_cache_get(pages[i]);
155                 }
156                 if (vmas)
157                         vmas[i] = vma;
158                 start += PAGE_SIZE;
159         }
160
161         return i;
162
163 finish_or_fault:
164         return i ? : -EFAULT;
165 }
166 EXPORT_SYMBOL(get_user_pages);
167
168 DEFINE_RWLOCK(vmlist_lock);
169 struct vm_struct *vmlist;
170
171 void vfree(const void *addr)
172 {
173         kfree(addr);
174 }
175 EXPORT_SYMBOL(vfree);
176
177 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
178 {
179         /*
180          *  You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
181          * returns only a logical address.
182          */
183         return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
184 }
185 EXPORT_SYMBOL(__vmalloc);
186
187 void *vmalloc_user(unsigned long size)
188 {
189         void *ret;
190
191         ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
192                         PAGE_KERNEL);
193         if (ret) {
194                 struct vm_area_struct *vma;
195
196                 down_write(&current->mm->mmap_sem);
197                 vma = find_vma(current->mm, (unsigned long)ret);
198                 if (vma)
199                         vma->vm_flags |= VM_USERMAP;
200                 up_write(&current->mm->mmap_sem);
201         }
202
203         return ret;
204 }
205 EXPORT_SYMBOL(vmalloc_user);
206
207 struct page *vmalloc_to_page(const void *addr)
208 {
209         return virt_to_page(addr);
210 }
211 EXPORT_SYMBOL(vmalloc_to_page);
212
213 unsigned long vmalloc_to_pfn(const void *addr)
214 {
215         return page_to_pfn(virt_to_page(addr));
216 }
217 EXPORT_SYMBOL(vmalloc_to_pfn);
218
219 long vread(char *buf, char *addr, unsigned long count)
220 {
221         memcpy(buf, addr, count);
222         return count;
223 }
224
225 long vwrite(char *buf, char *addr, unsigned long count)
226 {
227         /* Don't allow overflow */
228         if ((unsigned long) addr + count < count)
229                 count = -(unsigned long) addr;
230
231         memcpy(addr, buf, count);
232         return(count);
233 }
234
235 /*
236  *      vmalloc  -  allocate virtually continguos memory
237  *
238  *      @size:          allocation size
239  *
240  *      Allocate enough pages to cover @size from the page level
241  *      allocator and map them into continguos kernel virtual space.
242  *
243  *      For tight control over page level allocator and protection flags
244  *      use __vmalloc() instead.
245  */
246 void *vmalloc(unsigned long size)
247 {
248        return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
249 }
250 EXPORT_SYMBOL(vmalloc);
251
252 void *vmalloc_node(unsigned long size, int node)
253 {
254         return vmalloc(size);
255 }
256 EXPORT_SYMBOL(vmalloc_node);
257
258 /**
259  * vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
260  *      @size:          allocation size
261  *
262  *      Allocate enough 32bit PA addressable pages to cover @size from the
263  *      page level allocator and map them into continguos kernel virtual space.
264  */
265 void *vmalloc_32(unsigned long size)
266 {
267         return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
268 }
269 EXPORT_SYMBOL(vmalloc_32);
270
271 /**
272  * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
273  *      @size:          allocation size
274  *
275  * The resulting memory area is 32bit addressable and zeroed so it can be
276  * mapped to userspace without leaking data.
277  *
278  * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
279  * remap_vmalloc_range() are permissible.
280  */
281 void *vmalloc_32_user(unsigned long size)
282 {
283         /*
284          * We'll have to sort out the ZONE_DMA bits for 64-bit,
285          * but for now this can simply use vmalloc_user() directly.
286          */
287         return vmalloc_user(size);
288 }
289 EXPORT_SYMBOL(vmalloc_32_user);
290
291 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
292 {
293         BUG();
294         return NULL;
295 }
296 EXPORT_SYMBOL(vmap);
297
298 void vunmap(const void *addr)
299 {
300         BUG();
301 }
302 EXPORT_SYMBOL(vunmap);
303
304 /*
305  * Implement a stub for vmalloc_sync_all() if the architecture chose not to
306  * have one.
307  */
308 void  __attribute__((weak)) vmalloc_sync_all(void)
309 {
310 }
311
312 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
313                    struct page *page)
314 {
315         return -EINVAL;
316 }
317 EXPORT_SYMBOL(vm_insert_page);
318
319 /*
320  *  sys_brk() for the most part doesn't need the global kernel
321  *  lock, except when an application is doing something nasty
322  *  like trying to un-brk an area that has already been mapped
323  *  to a regular file.  in this case, the unmapping will need
324  *  to invoke file system routines that need the global lock.
325  */
326 asmlinkage unsigned long sys_brk(unsigned long brk)
327 {
328         struct mm_struct *mm = current->mm;
329
330         if (brk < mm->start_brk || brk > mm->context.end_brk)
331                 return mm->brk;
332
333         if (mm->brk == brk)
334                 return mm->brk;
335
336         /*
337          * Always allow shrinking brk
338          */
339         if (brk <= mm->brk) {
340                 mm->brk = brk;
341                 return brk;
342         }
343
344         /*
345          * Ok, looks good - let it rip.
346          */
347         return mm->brk = brk;
348 }
349
350 #ifdef DEBUG
351 static void show_process_blocks(void)
352 {
353         struct vm_list_struct *vml;
354
355         printk("Process blocks %d:", current->pid);
356
357         for (vml = &current->mm->context.vmlist; vml; vml = vml->next) {
358                 printk(" %p: %p", vml, vml->vma);
359                 if (vml->vma)
360                         printk(" (%d @%lx #%d)",
361                                kobjsize((void *) vml->vma->vm_start),
362                                vml->vma->vm_start,
363                                atomic_read(&vml->vma->vm_usage));
364                 printk(vml->next ? " ->" : ".\n");
365         }
366 }
367 #endif /* DEBUG */
368
369 /*
370  * add a VMA into a process's mm_struct in the appropriate place in the list
371  * - should be called with mm->mmap_sem held writelocked
372  */
373 static void add_vma_to_mm(struct mm_struct *mm, struct vm_list_struct *vml)
374 {
375         struct vm_list_struct **ppv;
376
377         for (ppv = &current->mm->context.vmlist; *ppv; ppv = &(*ppv)->next)
378                 if ((*ppv)->vma->vm_start > vml->vma->vm_start)
379                         break;
380
381         vml->next = *ppv;
382         *ppv = vml;
383 }
384
385 /*
386  * look up the first VMA in which addr resides, NULL if none
387  * - should be called with mm->mmap_sem at least held readlocked
388  */
389 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
390 {
391         struct vm_list_struct *loop, *vml;
392
393         /* search the vm_start ordered list */
394         vml = NULL;
395         for (loop = mm->context.vmlist; loop; loop = loop->next) {
396                 if (loop->vma->vm_start > addr)
397                         break;
398                 vml = loop;
399         }
400
401         if (vml && vml->vma->vm_end > addr)
402                 return vml->vma;
403
404         return NULL;
405 }
406 EXPORT_SYMBOL(find_vma);
407
408 /*
409  * find a VMA
410  * - we don't extend stack VMAs under NOMMU conditions
411  */
412 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
413 {
414         return find_vma(mm, addr);
415 }
416
417 int expand_stack(struct vm_area_struct *vma, unsigned long address)
418 {
419         return -ENOMEM;
420 }
421
422 /*
423  * look up the first VMA exactly that exactly matches addr
424  * - should be called with mm->mmap_sem at least held readlocked
425  */
426 static inline struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
427                                                     unsigned long addr)
428 {
429         struct vm_list_struct *vml;
430
431         /* search the vm_start ordered list */
432         for (vml = mm->context.vmlist; vml; vml = vml->next) {
433                 if (vml->vma->vm_start == addr)
434                         return vml->vma;
435                 if (vml->vma->vm_start > addr)
436                         break;
437         }
438
439         return NULL;
440 }
441
442 /*
443  * find a VMA in the global tree
444  */
445 static inline struct vm_area_struct *find_nommu_vma(unsigned long start)
446 {
447         struct vm_area_struct *vma;
448         struct rb_node *n = nommu_vma_tree.rb_node;
449
450         while (n) {
451                 vma = rb_entry(n, struct vm_area_struct, vm_rb);
452
453                 if (start < vma->vm_start)
454                         n = n->rb_left;
455                 else if (start > vma->vm_start)
456                         n = n->rb_right;
457                 else
458                         return vma;
459         }
460
461         return NULL;
462 }
463
464 /*
465  * add a VMA in the global tree
466  */
467 static void add_nommu_vma(struct vm_area_struct *vma)
468 {
469         struct vm_area_struct *pvma;
470         struct address_space *mapping;
471         struct rb_node **p = &nommu_vma_tree.rb_node;
472         struct rb_node *parent = NULL;
473
474         /* add the VMA to the mapping */
475         if (vma->vm_file) {
476                 mapping = vma->vm_file->f_mapping;
477
478                 flush_dcache_mmap_lock(mapping);
479                 vma_prio_tree_insert(vma, &mapping->i_mmap);
480                 flush_dcache_mmap_unlock(mapping);
481         }
482
483         /* add the VMA to the master list */
484         while (*p) {
485                 parent = *p;
486                 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
487
488                 if (vma->vm_start < pvma->vm_start) {
489                         p = &(*p)->rb_left;
490                 }
491                 else if (vma->vm_start > pvma->vm_start) {
492                         p = &(*p)->rb_right;
493                 }
494                 else {
495                         /* mappings are at the same address - this can only
496                          * happen for shared-mem chardevs and shared file
497                          * mappings backed by ramfs/tmpfs */
498                         BUG_ON(!(pvma->vm_flags & VM_SHARED));
499
500                         if (vma < pvma)
501                                 p = &(*p)->rb_left;
502                         else if (vma > pvma)
503                                 p = &(*p)->rb_right;
504                         else
505                                 BUG();
506                 }
507         }
508
509         rb_link_node(&vma->vm_rb, parent, p);
510         rb_insert_color(&vma->vm_rb, &nommu_vma_tree);
511 }
512
513 /*
514  * delete a VMA from the global list
515  */
516 static void delete_nommu_vma(struct vm_area_struct *vma)
517 {
518         struct address_space *mapping;
519
520         /* remove the VMA from the mapping */
521         if (vma->vm_file) {
522                 mapping = vma->vm_file->f_mapping;
523
524                 flush_dcache_mmap_lock(mapping);
525                 vma_prio_tree_remove(vma, &mapping->i_mmap);
526                 flush_dcache_mmap_unlock(mapping);
527         }
528
529         /* remove from the master list */
530         rb_erase(&vma->vm_rb, &nommu_vma_tree);
531 }
532
533 /*
534  * determine whether a mapping should be permitted and, if so, what sort of
535  * mapping we're capable of supporting
536  */
537 static int validate_mmap_request(struct file *file,
538                                  unsigned long addr,
539                                  unsigned long len,
540                                  unsigned long prot,
541                                  unsigned long flags,
542                                  unsigned long pgoff,
543                                  unsigned long *_capabilities)
544 {
545         unsigned long capabilities;
546         unsigned long reqprot = prot;
547         int ret;
548
549         /* do the simple checks first */
550         if (flags & MAP_FIXED || addr) {
551                 printk(KERN_DEBUG
552                        "%d: Can't do fixed-address/overlay mmap of RAM\n",
553                        current->pid);
554                 return -EINVAL;
555         }
556
557         if ((flags & MAP_TYPE) != MAP_PRIVATE &&
558             (flags & MAP_TYPE) != MAP_SHARED)
559                 return -EINVAL;
560
561         if (!len)
562                 return -EINVAL;
563
564         /* Careful about overflows.. */
565         len = PAGE_ALIGN(len);
566         if (!len || len > TASK_SIZE)
567                 return -ENOMEM;
568
569         /* offset overflow? */
570         if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
571                 return -EOVERFLOW;
572
573         if (file) {
574                 /* validate file mapping requests */
575                 struct address_space *mapping;
576
577                 /* files must support mmap */
578                 if (!file->f_op || !file->f_op->mmap)
579                         return -ENODEV;
580
581                 /* work out if what we've got could possibly be shared
582                  * - we support chardevs that provide their own "memory"
583                  * - we support files/blockdevs that are memory backed
584                  */
585                 mapping = file->f_mapping;
586                 if (!mapping)
587                         mapping = file->f_path.dentry->d_inode->i_mapping;
588
589                 capabilities = 0;
590                 if (mapping && mapping->backing_dev_info)
591                         capabilities = mapping->backing_dev_info->capabilities;
592
593                 if (!capabilities) {
594                         /* no explicit capabilities set, so assume some
595                          * defaults */
596                         switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) {
597                         case S_IFREG:
598                         case S_IFBLK:
599                                 capabilities = BDI_CAP_MAP_COPY;
600                                 break;
601
602                         case S_IFCHR:
603                                 capabilities =
604                                         BDI_CAP_MAP_DIRECT |
605                                         BDI_CAP_READ_MAP |
606                                         BDI_CAP_WRITE_MAP;
607                                 break;
608
609                         default:
610                                 return -EINVAL;
611                         }
612                 }
613
614                 /* eliminate any capabilities that we can't support on this
615                  * device */
616                 if (!file->f_op->get_unmapped_area)
617                         capabilities &= ~BDI_CAP_MAP_DIRECT;
618                 if (!file->f_op->read)
619                         capabilities &= ~BDI_CAP_MAP_COPY;
620
621                 if (flags & MAP_SHARED) {
622                         /* do checks for writing, appending and locking */
623                         if ((prot & PROT_WRITE) &&
624                             !(file->f_mode & FMODE_WRITE))
625                                 return -EACCES;
626
627                         if (IS_APPEND(file->f_path.dentry->d_inode) &&
628                             (file->f_mode & FMODE_WRITE))
629                                 return -EACCES;
630
631                         if (locks_verify_locked(file->f_path.dentry->d_inode))
632                                 return -EAGAIN;
633
634                         if (!(capabilities & BDI_CAP_MAP_DIRECT))
635                                 return -ENODEV;
636
637                         if (((prot & PROT_READ)  && !(capabilities & BDI_CAP_READ_MAP))  ||
638                             ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
639                             ((prot & PROT_EXEC)  && !(capabilities & BDI_CAP_EXEC_MAP))
640                             ) {
641                                 printk("MAP_SHARED not completely supported on !MMU\n");
642                                 return -EINVAL;
643                         }
644
645                         /* we mustn't privatise shared mappings */
646                         capabilities &= ~BDI_CAP_MAP_COPY;
647                 }
648                 else {
649                         /* we're going to read the file into private memory we
650                          * allocate */
651                         if (!(capabilities & BDI_CAP_MAP_COPY))
652                                 return -ENODEV;
653
654                         /* we don't permit a private writable mapping to be
655                          * shared with the backing device */
656                         if (prot & PROT_WRITE)
657                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
658                 }
659
660                 /* handle executable mappings and implied executable
661                  * mappings */
662                 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
663                         if (prot & PROT_EXEC)
664                                 return -EPERM;
665                 }
666                 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
667                         /* handle implication of PROT_EXEC by PROT_READ */
668                         if (current->personality & READ_IMPLIES_EXEC) {
669                                 if (capabilities & BDI_CAP_EXEC_MAP)
670                                         prot |= PROT_EXEC;
671                         }
672                 }
673                 else if ((prot & PROT_READ) &&
674                          (prot & PROT_EXEC) &&
675                          !(capabilities & BDI_CAP_EXEC_MAP)
676                          ) {
677                         /* backing file is not executable, try to copy */
678                         capabilities &= ~BDI_CAP_MAP_DIRECT;
679                 }
680         }
681         else {
682                 /* anonymous mappings are always memory backed and can be
683                  * privately mapped
684                  */
685                 capabilities = BDI_CAP_MAP_COPY;
686
687                 /* handle PROT_EXEC implication by PROT_READ */
688                 if ((prot & PROT_READ) &&
689                     (current->personality & READ_IMPLIES_EXEC))
690                         prot |= PROT_EXEC;
691         }
692
693         /* allow the security API to have its say */
694         ret = security_file_mmap(file, reqprot, prot, flags, addr, 0);
695         if (ret < 0)
696                 return ret;
697
698         /* looks okay */
699         *_capabilities = capabilities;
700         return 0;
701 }
702
703 /*
704  * we've determined that we can make the mapping, now translate what we
705  * now know into VMA flags
706  */
707 static unsigned long determine_vm_flags(struct file *file,
708                                         unsigned long prot,
709                                         unsigned long flags,
710                                         unsigned long capabilities)
711 {
712         unsigned long vm_flags;
713
714         vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
715         vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
716         /* vm_flags |= mm->def_flags; */
717
718         if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
719                 /* attempt to share read-only copies of mapped file chunks */
720                 if (file && !(prot & PROT_WRITE))
721                         vm_flags |= VM_MAYSHARE;
722         }
723         else {
724                 /* overlay a shareable mapping on the backing device or inode
725                  * if possible - used for chardevs, ramfs/tmpfs/shmfs and
726                  * romfs/cramfs */
727                 if (flags & MAP_SHARED)
728                         vm_flags |= VM_MAYSHARE | VM_SHARED;
729                 else if ((((vm_flags & capabilities) ^ vm_flags) & BDI_CAP_VMFLAGS) == 0)
730                         vm_flags |= VM_MAYSHARE;
731         }
732
733         /* refuse to let anyone share private mappings with this process if
734          * it's being traced - otherwise breakpoints set in it may interfere
735          * with another untraced process
736          */
737         if ((flags & MAP_PRIVATE) && (current->ptrace & PT_PTRACED))
738                 vm_flags &= ~VM_MAYSHARE;
739
740         return vm_flags;
741 }
742
743 /*
744  * set up a shared mapping on a file
745  */
746 static int do_mmap_shared_file(struct vm_area_struct *vma, unsigned long len)
747 {
748         int ret;
749
750         ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
751         if (ret != -ENOSYS)
752                 return ret;
753
754         /* getting an ENOSYS error indicates that direct mmap isn't
755          * possible (as opposed to tried but failed) so we'll fall
756          * through to making a private copy of the data and mapping
757          * that if we can */
758         return -ENODEV;
759 }
760
761 /*
762  * set up a private mapping or an anonymous shared mapping
763  */
764 static int do_mmap_private(struct vm_area_struct *vma, unsigned long len)
765 {
766         void *base;
767         int ret;
768
769         /* invoke the file's mapping function so that it can keep track of
770          * shared mappings on devices or memory
771          * - VM_MAYSHARE will be set if it may attempt to share
772          */
773         if (vma->vm_file) {
774                 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
775                 if (ret != -ENOSYS) {
776                         /* shouldn't return success if we're not sharing */
777                         BUG_ON(ret == 0 && !(vma->vm_flags & VM_MAYSHARE));
778                         return ret; /* success or a real error */
779                 }
780
781                 /* getting an ENOSYS error indicates that direct mmap isn't
782                  * possible (as opposed to tried but failed) so we'll try to
783                  * make a private copy of the data and map that instead */
784         }
785
786         /* allocate some memory to hold the mapping
787          * - note that this may not return a page-aligned address if the object
788          *   we're allocating is smaller than a page
789          */
790         base = kmalloc(len, GFP_KERNEL|__GFP_COMP);
791         if (!base)
792                 goto enomem;
793
794         vma->vm_start = (unsigned long) base;
795         vma->vm_end = vma->vm_start + len;
796         vma->vm_flags |= VM_MAPPED_COPY;
797
798 #ifdef WARN_ON_SLACK
799         if (len + WARN_ON_SLACK <= kobjsize(result))
800                 printk("Allocation of %lu bytes from process %d has %lu bytes of slack\n",
801                        len, current->pid, kobjsize(result) - len);
802 #endif
803
804         if (vma->vm_file) {
805                 /* read the contents of a file into the copy */
806                 mm_segment_t old_fs;
807                 loff_t fpos;
808
809                 fpos = vma->vm_pgoff;
810                 fpos <<= PAGE_SHIFT;
811
812                 old_fs = get_fs();
813                 set_fs(KERNEL_DS);
814                 ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos);
815                 set_fs(old_fs);
816
817                 if (ret < 0)
818                         goto error_free;
819
820                 /* clear the last little bit */
821                 if (ret < len)
822                         memset(base + ret, 0, len - ret);
823
824         } else {
825                 /* if it's an anonymous mapping, then just clear it */
826                 memset(base, 0, len);
827         }
828
829         return 0;
830
831 error_free:
832         kfree(base);
833         vma->vm_start = 0;
834         return ret;
835
836 enomem:
837         printk("Allocation of length %lu from process %d failed\n",
838                len, current->pid);
839         show_free_areas();
840         return -ENOMEM;
841 }
842
843 /*
844  * handle mapping creation for uClinux
845  */
846 unsigned long do_mmap_pgoff(struct file *file,
847                             unsigned long addr,
848                             unsigned long len,
849                             unsigned long prot,
850                             unsigned long flags,
851                             unsigned long pgoff)
852 {
853         struct vm_list_struct *vml = NULL;
854         struct vm_area_struct *vma = NULL;
855         struct rb_node *rb;
856         unsigned long capabilities, vm_flags;
857         void *result;
858         int ret;
859
860         if (!(flags & MAP_FIXED))
861                 addr = round_hint_to_min(addr);
862
863         /* decide whether we should attempt the mapping, and if so what sort of
864          * mapping */
865         ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
866                                     &capabilities);
867         if (ret < 0)
868                 return ret;
869
870         /* we've determined that we can make the mapping, now translate what we
871          * now know into VMA flags */
872         vm_flags = determine_vm_flags(file, prot, flags, capabilities);
873
874         /* we're going to need to record the mapping if it works */
875         vml = kzalloc(sizeof(struct vm_list_struct), GFP_KERNEL);
876         if (!vml)
877                 goto error_getting_vml;
878
879         down_write(&nommu_vma_sem);
880
881         /* if we want to share, we need to check for VMAs created by other
882          * mmap() calls that overlap with our proposed mapping
883          * - we can only share with an exact match on most regular files
884          * - shared mappings on character devices and memory backed files are
885          *   permitted to overlap inexactly as far as we are concerned for in
886          *   these cases, sharing is handled in the driver or filesystem rather
887          *   than here
888          */
889         if (vm_flags & VM_MAYSHARE) {
890                 unsigned long pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
891                 unsigned long vmpglen;
892
893                 /* suppress VMA sharing for shared regions */
894                 if (vm_flags & VM_SHARED &&
895                     capabilities & BDI_CAP_MAP_DIRECT)
896                         goto dont_share_VMAs;
897
898                 for (rb = rb_first(&nommu_vma_tree); rb; rb = rb_next(rb)) {
899                         vma = rb_entry(rb, struct vm_area_struct, vm_rb);
900
901                         if (!(vma->vm_flags & VM_MAYSHARE))
902                                 continue;
903
904                         /* search for overlapping mappings on the same file */
905                         if (vma->vm_file->f_path.dentry->d_inode != file->f_path.dentry->d_inode)
906                                 continue;
907
908                         if (vma->vm_pgoff >= pgoff + pglen)
909                                 continue;
910
911                         vmpglen = vma->vm_end - vma->vm_start + PAGE_SIZE - 1;
912                         vmpglen >>= PAGE_SHIFT;
913                         if (pgoff >= vma->vm_pgoff + vmpglen)
914                                 continue;
915
916                         /* handle inexactly overlapping matches between mappings */
917                         if (vma->vm_pgoff != pgoff || vmpglen != pglen) {
918                                 if (!(capabilities & BDI_CAP_MAP_DIRECT))
919                                         goto sharing_violation;
920                                 continue;
921                         }
922
923                         /* we've found a VMA we can share */
924                         atomic_inc(&vma->vm_usage);
925
926                         vml->vma = vma;
927                         result = (void *) vma->vm_start;
928                         goto shared;
929                 }
930
931         dont_share_VMAs:
932                 vma = NULL;
933
934                 /* obtain the address at which to make a shared mapping
935                  * - this is the hook for quasi-memory character devices to
936                  *   tell us the location of a shared mapping
937                  */
938                 if (file && file->f_op->get_unmapped_area) {
939                         addr = file->f_op->get_unmapped_area(file, addr, len,
940                                                              pgoff, flags);
941                         if (IS_ERR((void *) addr)) {
942                                 ret = addr;
943                                 if (ret != (unsigned long) -ENOSYS)
944                                         goto error;
945
946                                 /* the driver refused to tell us where to site
947                                  * the mapping so we'll have to attempt to copy
948                                  * it */
949                                 ret = (unsigned long) -ENODEV;
950                                 if (!(capabilities & BDI_CAP_MAP_COPY))
951                                         goto error;
952
953                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
954                         }
955                 }
956         }
957
958         /* we're going to need a VMA struct as well */
959         vma = kzalloc(sizeof(struct vm_area_struct), GFP_KERNEL);
960         if (!vma)
961                 goto error_getting_vma;
962
963         INIT_LIST_HEAD(&vma->anon_vma_node);
964         atomic_set(&vma->vm_usage, 1);
965         if (file)
966                 get_file(file);
967         vma->vm_file    = file;
968         vma->vm_flags   = vm_flags;
969         vma->vm_start   = addr;
970         vma->vm_end     = addr + len;
971         vma->vm_pgoff   = pgoff;
972
973         vml->vma = vma;
974
975         /* set up the mapping */
976         if (file && vma->vm_flags & VM_SHARED)
977                 ret = do_mmap_shared_file(vma, len);
978         else
979                 ret = do_mmap_private(vma, len);
980         if (ret < 0)
981                 goto error;
982
983         /* okay... we have a mapping; now we have to register it */
984         result = (void *) vma->vm_start;
985
986         if (vma->vm_flags & VM_MAPPED_COPY) {
987                 realalloc += kobjsize(result);
988                 askedalloc += len;
989         }
990
991         realalloc += kobjsize(vma);
992         askedalloc += sizeof(*vma);
993
994         current->mm->total_vm += len >> PAGE_SHIFT;
995
996         add_nommu_vma(vma);
997
998  shared:
999         realalloc += kobjsize(vml);
1000         askedalloc += sizeof(*vml);
1001
1002         add_vma_to_mm(current->mm, vml);
1003
1004         up_write(&nommu_vma_sem);
1005
1006         if (prot & PROT_EXEC)
1007                 flush_icache_range((unsigned long) result,
1008                                    (unsigned long) result + len);
1009
1010 #ifdef DEBUG
1011         printk("do_mmap:\n");
1012         show_process_blocks();
1013 #endif
1014
1015         return (unsigned long) result;
1016
1017  error:
1018         up_write(&nommu_vma_sem);
1019         kfree(vml);
1020         if (vma) {
1021                 if (vma->vm_file)
1022                         fput(vma->vm_file);
1023                 kfree(vma);
1024         }
1025         return ret;
1026
1027  sharing_violation:
1028         up_write(&nommu_vma_sem);
1029         printk("Attempt to share mismatched mappings\n");
1030         kfree(vml);
1031         return -EINVAL;
1032
1033  error_getting_vma:
1034         up_write(&nommu_vma_sem);
1035         kfree(vml);
1036         printk("Allocation of vma for %lu byte allocation from process %d failed\n",
1037                len, current->pid);
1038         show_free_areas();
1039         return -ENOMEM;
1040
1041  error_getting_vml:
1042         printk("Allocation of vml for %lu byte allocation from process %d failed\n",
1043                len, current->pid);
1044         show_free_areas();
1045         return -ENOMEM;
1046 }
1047 EXPORT_SYMBOL(do_mmap_pgoff);
1048
1049 /*
1050  * handle mapping disposal for uClinux
1051  */
1052 static void put_vma(struct vm_area_struct *vma)
1053 {
1054         if (vma) {
1055                 down_write(&nommu_vma_sem);
1056
1057                 if (atomic_dec_and_test(&vma->vm_usage)) {
1058                         delete_nommu_vma(vma);
1059
1060                         if (vma->vm_ops && vma->vm_ops->close)
1061                                 vma->vm_ops->close(vma);
1062
1063                         /* IO memory and memory shared directly out of the pagecache from
1064                          * ramfs/tmpfs mustn't be released here */
1065                         if (vma->vm_flags & VM_MAPPED_COPY) {
1066                                 realalloc -= kobjsize((void *) vma->vm_start);
1067                                 askedalloc -= vma->vm_end - vma->vm_start;
1068                                 kfree((void *) vma->vm_start);
1069                         }
1070
1071                         realalloc -= kobjsize(vma);
1072                         askedalloc -= sizeof(*vma);
1073
1074                         if (vma->vm_file)
1075                                 fput(vma->vm_file);
1076                         kfree(vma);
1077                 }
1078
1079                 up_write(&nommu_vma_sem);
1080         }
1081 }
1082
1083 /*
1084  * release a mapping
1085  * - under NOMMU conditions the parameters must match exactly to the mapping to
1086  *   be removed
1087  */
1088 int do_munmap(struct mm_struct *mm, unsigned long addr, size_t len)
1089 {
1090         struct vm_list_struct *vml, **parent;
1091         unsigned long end = addr + len;
1092
1093 #ifdef DEBUG
1094         printk("do_munmap:\n");
1095 #endif
1096
1097         for (parent = &mm->context.vmlist; *parent; parent = &(*parent)->next) {
1098                 if ((*parent)->vma->vm_start > addr)
1099                         break;
1100                 if ((*parent)->vma->vm_start == addr &&
1101                     ((len == 0) || ((*parent)->vma->vm_end == end)))
1102                         goto found;
1103         }
1104
1105         printk("munmap of non-mmaped memory by process %d (%s): %p\n",
1106                current->pid, current->comm, (void *) addr);
1107         return -EINVAL;
1108
1109  found:
1110         vml = *parent;
1111
1112         put_vma(vml->vma);
1113
1114         *parent = vml->next;
1115         realalloc -= kobjsize(vml);
1116         askedalloc -= sizeof(*vml);
1117         kfree(vml);
1118
1119         update_hiwater_vm(mm);
1120         mm->total_vm -= len >> PAGE_SHIFT;
1121
1122 #ifdef DEBUG
1123         show_process_blocks();
1124 #endif
1125
1126         return 0;
1127 }
1128 EXPORT_SYMBOL(do_munmap);
1129
1130 asmlinkage long sys_munmap(unsigned long addr, size_t len)
1131 {
1132         int ret;
1133         struct mm_struct *mm = current->mm;
1134
1135         down_write(&mm->mmap_sem);
1136         ret = do_munmap(mm, addr, len);
1137         up_write(&mm->mmap_sem);
1138         return ret;
1139 }
1140
1141 /*
1142  * Release all mappings
1143  */
1144 void exit_mmap(struct mm_struct * mm)
1145 {
1146         struct vm_list_struct *tmp;
1147
1148         if (mm) {
1149 #ifdef DEBUG
1150                 printk("Exit_mmap:\n");
1151 #endif
1152
1153                 mm->total_vm = 0;
1154
1155                 while ((tmp = mm->context.vmlist)) {
1156                         mm->context.vmlist = tmp->next;
1157                         put_vma(tmp->vma);
1158
1159                         realalloc -= kobjsize(tmp);
1160                         askedalloc -= sizeof(*tmp);
1161                         kfree(tmp);
1162                 }
1163
1164 #ifdef DEBUG
1165                 show_process_blocks();
1166 #endif
1167         }
1168 }
1169
1170 unsigned long do_brk(unsigned long addr, unsigned long len)
1171 {
1172         return -ENOMEM;
1173 }
1174
1175 /*
1176  * expand (or shrink) an existing mapping, potentially moving it at the same
1177  * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1178  *
1179  * under NOMMU conditions, we only permit changing a mapping's size, and only
1180  * as long as it stays within the hole allocated by the kmalloc() call in
1181  * do_mmap_pgoff() and the block is not shareable
1182  *
1183  * MREMAP_FIXED is not supported under NOMMU conditions
1184  */
1185 unsigned long do_mremap(unsigned long addr,
1186                         unsigned long old_len, unsigned long new_len,
1187                         unsigned long flags, unsigned long new_addr)
1188 {
1189         struct vm_area_struct *vma;
1190
1191         /* insanity checks first */
1192         if (new_len == 0)
1193                 return (unsigned long) -EINVAL;
1194
1195         if (flags & MREMAP_FIXED && new_addr != addr)
1196                 return (unsigned long) -EINVAL;
1197
1198         vma = find_vma_exact(current->mm, addr);
1199         if (!vma)
1200                 return (unsigned long) -EINVAL;
1201
1202         if (vma->vm_end != vma->vm_start + old_len)
1203                 return (unsigned long) -EFAULT;
1204
1205         if (vma->vm_flags & VM_MAYSHARE)
1206                 return (unsigned long) -EPERM;
1207
1208         if (new_len > kobjsize((void *) addr))
1209                 return (unsigned long) -ENOMEM;
1210
1211         /* all checks complete - do it */
1212         vma->vm_end = vma->vm_start + new_len;
1213
1214         askedalloc -= old_len;
1215         askedalloc += new_len;
1216
1217         return vma->vm_start;
1218 }
1219 EXPORT_SYMBOL(do_mremap);
1220
1221 asmlinkage unsigned long sys_mremap(unsigned long addr,
1222         unsigned long old_len, unsigned long new_len,
1223         unsigned long flags, unsigned long new_addr)
1224 {
1225         unsigned long ret;
1226
1227         down_write(&current->mm->mmap_sem);
1228         ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1229         up_write(&current->mm->mmap_sem);
1230         return ret;
1231 }
1232
1233 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1234                         unsigned int foll_flags)
1235 {
1236         return NULL;
1237 }
1238
1239 int remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
1240                 unsigned long to, unsigned long size, pgprot_t prot)
1241 {
1242         vma->vm_start = vma->vm_pgoff << PAGE_SHIFT;
1243         return 0;
1244 }
1245 EXPORT_SYMBOL(remap_pfn_range);
1246
1247 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1248                         unsigned long pgoff)
1249 {
1250         unsigned int size = vma->vm_end - vma->vm_start;
1251
1252         if (!(vma->vm_flags & VM_USERMAP))
1253                 return -EINVAL;
1254
1255         vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1256         vma->vm_end = vma->vm_start + size;
1257
1258         return 0;
1259 }
1260 EXPORT_SYMBOL(remap_vmalloc_range);
1261
1262 void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1263 {
1264 }
1265
1266 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1267         unsigned long len, unsigned long pgoff, unsigned long flags)
1268 {
1269         return -ENOMEM;
1270 }
1271
1272 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1273 {
1274 }
1275
1276 void unmap_mapping_range(struct address_space *mapping,
1277                          loff_t const holebegin, loff_t const holelen,
1278                          int even_cows)
1279 {
1280 }
1281 EXPORT_SYMBOL(unmap_mapping_range);
1282
1283 /*
1284  * ask for an unmapped area at which to create a mapping on a file
1285  */
1286 unsigned long get_unmapped_area(struct file *file, unsigned long addr,
1287                                 unsigned long len, unsigned long pgoff,
1288                                 unsigned long flags)
1289 {
1290         unsigned long (*get_area)(struct file *, unsigned long, unsigned long,
1291                                   unsigned long, unsigned long);
1292
1293         get_area = current->mm->get_unmapped_area;
1294         if (file && file->f_op && file->f_op->get_unmapped_area)
1295                 get_area = file->f_op->get_unmapped_area;
1296
1297         if (!get_area)
1298                 return -ENOSYS;
1299
1300         return get_area(file, addr, len, pgoff, flags);
1301 }
1302 EXPORT_SYMBOL(get_unmapped_area);
1303
1304 /*
1305  * Check that a process has enough memory to allocate a new virtual
1306  * mapping. 0 means there is enough memory for the allocation to
1307  * succeed and -ENOMEM implies there is not.
1308  *
1309  * We currently support three overcommit policies, which are set via the
1310  * vm.overcommit_memory sysctl.  See Documentation/vm/overcommit-accounting
1311  *
1312  * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1313  * Additional code 2002 Jul 20 by Robert Love.
1314  *
1315  * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1316  *
1317  * Note this is a helper function intended to be used by LSMs which
1318  * wish to use this logic.
1319  */
1320 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
1321 {
1322         unsigned long free, allowed;
1323
1324         vm_acct_memory(pages);
1325
1326         /*
1327          * Sometimes we want to use more memory than we have
1328          */
1329         if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1330                 return 0;
1331
1332         if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1333                 unsigned long n;
1334
1335                 free = global_page_state(NR_FILE_PAGES);
1336                 free += nr_swap_pages;
1337
1338                 /*
1339                  * Any slabs which are created with the
1340                  * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1341                  * which are reclaimable, under pressure.  The dentry
1342                  * cache and most inode caches should fall into this
1343                  */
1344                 free += global_page_state(NR_SLAB_RECLAIMABLE);
1345
1346                 /*
1347                  * Leave the last 3% for root
1348                  */
1349                 if (!cap_sys_admin)
1350                         free -= free / 32;
1351
1352                 if (free > pages)
1353                         return 0;
1354
1355                 /*
1356                  * nr_free_pages() is very expensive on large systems,
1357                  * only call if we're about to fail.
1358                  */
1359                 n = nr_free_pages();
1360
1361                 /*
1362                  * Leave reserved pages. The pages are not for anonymous pages.
1363                  */
1364                 if (n <= totalreserve_pages)
1365                         goto error;
1366                 else
1367                         n -= totalreserve_pages;
1368
1369                 /*
1370                  * Leave the last 3% for root
1371                  */
1372                 if (!cap_sys_admin)
1373                         n -= n / 32;
1374                 free += n;
1375
1376                 if (free > pages)
1377                         return 0;
1378
1379                 goto error;
1380         }
1381
1382         allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1383         /*
1384          * Leave the last 3% for root
1385          */
1386         if (!cap_sys_admin)
1387                 allowed -= allowed / 32;
1388         allowed += total_swap_pages;
1389
1390         /* Don't let a single process grow too big:
1391            leave 3% of the size of this process for other processes */
1392         allowed -= current->mm->total_vm / 32;
1393
1394         /*
1395          * cast `allowed' as a signed long because vm_committed_space
1396          * sometimes has a negative value
1397          */
1398         if (atomic_read(&vm_committed_space) < (long)allowed)
1399                 return 0;
1400 error:
1401         vm_unacct_memory(pages);
1402
1403         return -ENOMEM;
1404 }
1405
1406 int in_gate_area_no_task(unsigned long addr)
1407 {
1408         return 0;
1409 }
1410
1411 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1412 {
1413         BUG();
1414         return 0;
1415 }
1416 EXPORT_SYMBOL(filemap_fault);
1417
1418 /*
1419  * Access another process' address space.
1420  * - source/target buffer must be kernel space
1421  */
1422 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
1423 {
1424         struct vm_area_struct *vma;
1425         struct mm_struct *mm;
1426
1427         if (addr + len < addr)
1428                 return 0;
1429
1430         mm = get_task_mm(tsk);
1431         if (!mm)
1432                 return 0;
1433
1434         down_read(&mm->mmap_sem);
1435
1436         /* the access must start within one of the target process's mappings */
1437         vma = find_vma(mm, addr);
1438         if (vma) {
1439                 /* don't overrun this mapping */
1440                 if (addr + len >= vma->vm_end)
1441                         len = vma->vm_end - addr;
1442
1443                 /* only read or write mappings where it is permitted */
1444                 if (write && vma->vm_flags & VM_MAYWRITE)
1445                         len -= copy_to_user((void *) addr, buf, len);
1446                 else if (!write && vma->vm_flags & VM_MAYREAD)
1447                         len -= copy_from_user(buf, (void *) addr, len);
1448                 else
1449                         len = 0;
1450         } else {
1451                 len = 0;
1452         }
1453
1454         up_read(&mm->mmap_sem);
1455         mmput(mm);
1456         return len;
1457 }