[PATCH] NOMMU: Make mremap() partially work for NOMMU kernels
[linux-2.6.git] / mm / nommu.c
1 /*
2  *  linux/mm/nommu.c
3  *
4  *  Replacement code for mm functions to support CPU's that don't
5  *  have any form of memory management unit (thus no virtual memory).
6  *
7  *  See Documentation/nommu-mmap.txt
8  *
9  *  Copyright (c) 2004-2005 David Howells <dhowells@redhat.com>
10  *  Copyright (c) 2000-2003 David McCullough <davidm@snapgear.com>
11  *  Copyright (c) 2000-2001 D Jeff Dionne <jeff@uClinux.org>
12  *  Copyright (c) 2002      Greg Ungerer <gerg@snapgear.com>
13  */
14
15 #include <linux/mm.h>
16 #include <linux/mman.h>
17 #include <linux/swap.h>
18 #include <linux/file.h>
19 #include <linux/highmem.h>
20 #include <linux/pagemap.h>
21 #include <linux/slab.h>
22 #include <linux/vmalloc.h>
23 #include <linux/ptrace.h>
24 #include <linux/blkdev.h>
25 #include <linux/backing-dev.h>
26 #include <linux/mount.h>
27 #include <linux/personality.h>
28 #include <linux/security.h>
29 #include <linux/syscalls.h>
30
31 #include <asm/uaccess.h>
32 #include <asm/tlb.h>
33 #include <asm/tlbflush.h>
34
35 void *high_memory;
36 struct page *mem_map;
37 unsigned long max_mapnr;
38 unsigned long num_physpages;
39 unsigned long askedalloc, realalloc;
40 atomic_t vm_committed_space = ATOMIC_INIT(0);
41 int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
42 int sysctl_overcommit_ratio = 50; /* default is 50% */
43 int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
44 int heap_stack_gap = 0;
45
46 EXPORT_SYMBOL(mem_map);
47 EXPORT_SYMBOL(__vm_enough_memory);
48
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 EXPORT_SYMBOL(vfree);
57 EXPORT_SYMBOL(vmalloc_to_page);
58 EXPORT_SYMBOL(vmalloc_32);
59 EXPORT_SYMBOL(vmap);
60 EXPORT_SYMBOL(vunmap);
61
62 /*
63  * Handle all mappings that got truncated by a "truncate()"
64  * system call.
65  *
66  * NOTE! We have to be ready to update the memory sharing
67  * between the file and the memory map for a potential last
68  * incomplete page.  Ugly, but necessary.
69  */
70 int vmtruncate(struct inode *inode, loff_t offset)
71 {
72         struct address_space *mapping = inode->i_mapping;
73         unsigned long limit;
74
75         if (inode->i_size < offset)
76                 goto do_expand;
77         i_size_write(inode, offset);
78
79         truncate_inode_pages(mapping, offset);
80         goto out_truncate;
81
82 do_expand:
83         limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
84         if (limit != RLIM_INFINITY && offset > limit)
85                 goto out_sig;
86         if (offset > inode->i_sb->s_maxbytes)
87                 goto out;
88         i_size_write(inode, offset);
89
90 out_truncate:
91         if (inode->i_op && inode->i_op->truncate)
92                 inode->i_op->truncate(inode);
93         return 0;
94 out_sig:
95         send_sig(SIGXFSZ, current, 0);
96 out:
97         return -EFBIG;
98 }
99
100 EXPORT_SYMBOL(vmtruncate);
101
102 /*
103  * Return the total memory allocated for this pointer, not
104  * just what the caller asked for.
105  *
106  * Doesn't have to be accurate, i.e. may have races.
107  */
108 unsigned int kobjsize(const void *objp)
109 {
110         struct page *page;
111
112         if (!objp || !((page = virt_to_page(objp))))
113                 return 0;
114
115         if (PageSlab(page))
116                 return ksize(objp);
117
118         BUG_ON(page->index < 0);
119         BUG_ON(page->index >= MAX_ORDER);
120
121         return (PAGE_SIZE << page->index);
122 }
123
124 /*
125  * get a list of pages in an address range belonging to the specified process
126  * and indicate the VMA that covers each page
127  * - this is potentially dodgy as we may end incrementing the page count of a
128  *   slab page or a secondary page from a compound page
129  * - don't permit access to VMAs that don't support it, such as I/O mappings
130  */
131 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
132         unsigned long start, int len, int write, int force,
133         struct page **pages, struct vm_area_struct **vmas)
134 {
135         struct vm_area_struct *vma;
136         unsigned long vm_flags;
137         int i;
138
139         /* calculate required read or write permissions.
140          * - if 'force' is set, we only require the "MAY" flags.
141          */
142         vm_flags  = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
143         vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
144
145         for (i = 0; i < len; i++) {
146                 vma = find_vma(mm, start);
147                 if (!vma)
148                         goto finish_or_fault;
149
150                 /* protect what we can, including chardevs */
151                 if (vma->vm_flags & (VM_IO | VM_PFNMAP) ||
152                     !(vm_flags & vma->vm_flags))
153                         goto finish_or_fault;
154
155                 if (pages) {
156                         pages[i] = virt_to_page(start);
157                         if (pages[i])
158                                 page_cache_get(pages[i]);
159                 }
160                 if (vmas)
161                         vmas[i] = vma;
162                 start += PAGE_SIZE;
163         }
164
165         return i;
166
167 finish_or_fault:
168         return i ? : -EFAULT;
169 }
170
171 EXPORT_SYMBOL(get_user_pages);
172
173 DEFINE_RWLOCK(vmlist_lock);
174 struct vm_struct *vmlist;
175
176 void vfree(void *addr)
177 {
178         kfree(addr);
179 }
180
181 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
182 {
183         /*
184          * kmalloc doesn't like __GFP_HIGHMEM for some reason
185          */
186         return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
187 }
188
189 struct page * vmalloc_to_page(void *addr)
190 {
191         return virt_to_page(addr);
192 }
193
194 unsigned long vmalloc_to_pfn(void *addr)
195 {
196         return page_to_pfn(virt_to_page(addr));
197 }
198
199
200 long vread(char *buf, char *addr, unsigned long count)
201 {
202         memcpy(buf, addr, count);
203         return count;
204 }
205
206 long vwrite(char *buf, char *addr, unsigned long count)
207 {
208         /* Don't allow overflow */
209         if ((unsigned long) addr + count < count)
210                 count = -(unsigned long) addr;
211
212         memcpy(addr, buf, count);
213         return(count);
214 }
215
216 /*
217  *      vmalloc  -  allocate virtually continguos memory
218  *
219  *      @size:          allocation size
220  *
221  *      Allocate enough pages to cover @size from the page level
222  *      allocator and map them into continguos kernel virtual space.
223  *
224  *      For tight cotrol over page level allocator and protection flags
225  *      use __vmalloc() instead.
226  */
227 void *vmalloc(unsigned long size)
228 {
229        return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
230 }
231 EXPORT_SYMBOL(vmalloc);
232
233 void *vmalloc_node(unsigned long size, int node)
234 {
235         return vmalloc(size);
236 }
237 EXPORT_SYMBOL(vmalloc_node);
238
239 /*
240  *      vmalloc_32  -  allocate virtually continguos memory (32bit addressable)
241  *
242  *      @size:          allocation size
243  *
244  *      Allocate enough 32bit PA addressable pages to cover @size from the
245  *      page level allocator and map them into continguos kernel virtual space.
246  */
247 void *vmalloc_32(unsigned long size)
248 {
249         return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
250 }
251
252 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
253 {
254         BUG();
255         return NULL;
256 }
257
258 void vunmap(void *addr)
259 {
260         BUG();
261 }
262
263 /*
264  *  sys_brk() for the most part doesn't need the global kernel
265  *  lock, except when an application is doing something nasty
266  *  like trying to un-brk an area that has already been mapped
267  *  to a regular file.  in this case, the unmapping will need
268  *  to invoke file system routines that need the global lock.
269  */
270 asmlinkage unsigned long sys_brk(unsigned long brk)
271 {
272         struct mm_struct *mm = current->mm;
273
274         if (brk < mm->start_brk || brk > mm->context.end_brk)
275                 return mm->brk;
276
277         if (mm->brk == brk)
278                 return mm->brk;
279
280         /*
281          * Always allow shrinking brk
282          */
283         if (brk <= mm->brk) {
284                 mm->brk = brk;
285                 return brk;
286         }
287
288         /*
289          * Ok, looks good - let it rip.
290          */
291         return mm->brk = brk;
292 }
293
294 #ifdef DEBUG
295 static void show_process_blocks(void)
296 {
297         struct vm_list_struct *vml;
298
299         printk("Process blocks %d:", current->pid);
300
301         for (vml = &current->mm->context.vmlist; vml; vml = vml->next) {
302                 printk(" %p: %p", vml, vml->vma);
303                 if (vml->vma)
304                         printk(" (%d @%lx #%d)",
305                                kobjsize((void *) vml->vma->vm_start),
306                                vml->vma->vm_start,
307                                atomic_read(&vml->vma->vm_usage));
308                 printk(vml->next ? " ->" : ".\n");
309         }
310 }
311 #endif /* DEBUG */
312
313 /*
314  * add a VMA into a process's mm_struct in the appropriate place in the list
315  * - should be called with mm->mmap_sem held writelocked
316  */
317 static void add_vma_to_mm(struct mm_struct *mm, struct vm_list_struct *vml)
318 {
319         struct vm_list_struct **ppv;
320
321         for (ppv = &current->mm->context.vmlist; *ppv; ppv = &(*ppv)->next)
322                 if ((*ppv)->vma->vm_start > vml->vma->vm_start)
323                         break;
324
325         vml->next = *ppv;
326         *ppv = vml;
327 }
328
329 /*
330  * look up the first VMA in which addr resides, NULL if none
331  * - should be called with mm->mmap_sem at least held readlocked
332  */
333 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
334 {
335         struct vm_list_struct *loop, *vml;
336
337         /* search the vm_start ordered list */
338         vml = NULL;
339         for (loop = mm->context.vmlist; loop; loop = loop->next) {
340                 if (loop->vma->vm_start > addr)
341                         break;
342                 vml = loop;
343         }
344
345         if (vml && vml->vma->vm_end > addr)
346                 return vml->vma;
347
348         return NULL;
349 }
350 EXPORT_SYMBOL(find_vma);
351
352 /*
353  * look up the first VMA exactly that exactly matches addr
354  * - should be called with mm->mmap_sem at least held readlocked
355  */
356 static inline struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
357                                                     unsigned long addr)
358 {
359         struct vm_list_struct *vml;
360
361         /* search the vm_start ordered list */
362         for (vml = mm->context.vmlist; vml; vml = vml->next) {
363                 if (vml->vma->vm_start == addr)
364                         return vml->vma;
365                 if (vml->vma->vm_start > addr)
366                         break;
367         }
368
369         return NULL;
370 }
371
372 /*
373  * find a VMA in the global tree
374  */
375 static inline struct vm_area_struct *find_nommu_vma(unsigned long start)
376 {
377         struct vm_area_struct *vma;
378         struct rb_node *n = nommu_vma_tree.rb_node;
379
380         while (n) {
381                 vma = rb_entry(n, struct vm_area_struct, vm_rb);
382
383                 if (start < vma->vm_start)
384                         n = n->rb_left;
385                 else if (start > vma->vm_start)
386                         n = n->rb_right;
387                 else
388                         return vma;
389         }
390
391         return NULL;
392 }
393
394 /*
395  * add a VMA in the global tree
396  */
397 static void add_nommu_vma(struct vm_area_struct *vma)
398 {
399         struct vm_area_struct *pvma;
400         struct address_space *mapping;
401         struct rb_node **p = &nommu_vma_tree.rb_node;
402         struct rb_node *parent = NULL;
403
404         /* add the VMA to the mapping */
405         if (vma->vm_file) {
406                 mapping = vma->vm_file->f_mapping;
407
408                 flush_dcache_mmap_lock(mapping);
409                 vma_prio_tree_insert(vma, &mapping->i_mmap);
410                 flush_dcache_mmap_unlock(mapping);
411         }
412
413         /* add the VMA to the master list */
414         while (*p) {
415                 parent = *p;
416                 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
417
418                 if (vma->vm_start < pvma->vm_start) {
419                         p = &(*p)->rb_left;
420                 }
421                 else if (vma->vm_start > pvma->vm_start) {
422                         p = &(*p)->rb_right;
423                 }
424                 else {
425                         /* mappings are at the same address - this can only
426                          * happen for shared-mem chardevs and shared file
427                          * mappings backed by ramfs/tmpfs */
428                         BUG_ON(!(pvma->vm_flags & VM_SHARED));
429
430                         if (vma < pvma)
431                                 p = &(*p)->rb_left;
432                         else if (vma > pvma)
433                                 p = &(*p)->rb_right;
434                         else
435                                 BUG();
436                 }
437         }
438
439         rb_link_node(&vma->vm_rb, parent, p);
440         rb_insert_color(&vma->vm_rb, &nommu_vma_tree);
441 }
442
443 /*
444  * delete a VMA from the global list
445  */
446 static void delete_nommu_vma(struct vm_area_struct *vma)
447 {
448         struct address_space *mapping;
449
450         /* remove the VMA from the mapping */
451         if (vma->vm_file) {
452                 mapping = vma->vm_file->f_mapping;
453
454                 flush_dcache_mmap_lock(mapping);
455                 vma_prio_tree_remove(vma, &mapping->i_mmap);
456                 flush_dcache_mmap_unlock(mapping);
457         }
458
459         /* remove from the master list */
460         rb_erase(&vma->vm_rb, &nommu_vma_tree);
461 }
462
463 /*
464  * determine whether a mapping should be permitted and, if so, what sort of
465  * mapping we're capable of supporting
466  */
467 static int validate_mmap_request(struct file *file,
468                                  unsigned long addr,
469                                  unsigned long len,
470                                  unsigned long prot,
471                                  unsigned long flags,
472                                  unsigned long pgoff,
473                                  unsigned long *_capabilities)
474 {
475         unsigned long capabilities;
476         unsigned long reqprot = prot;
477         int ret;
478
479         /* do the simple checks first */
480         if (flags & MAP_FIXED || addr) {
481                 printk(KERN_DEBUG
482                        "%d: Can't do fixed-address/overlay mmap of RAM\n",
483                        current->pid);
484                 return -EINVAL;
485         }
486
487         if ((flags & MAP_TYPE) != MAP_PRIVATE &&
488             (flags & MAP_TYPE) != MAP_SHARED)
489                 return -EINVAL;
490
491         if (PAGE_ALIGN(len) == 0)
492                 return addr;
493
494         if (len > TASK_SIZE)
495                 return -EINVAL;
496
497         /* offset overflow? */
498         if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
499                 return -EINVAL;
500
501         if (file) {
502                 /* validate file mapping requests */
503                 struct address_space *mapping;
504
505                 /* files must support mmap */
506                 if (!file->f_op || !file->f_op->mmap)
507                         return -ENODEV;
508
509                 /* work out if what we've got could possibly be shared
510                  * - we support chardevs that provide their own "memory"
511                  * - we support files/blockdevs that are memory backed
512                  */
513                 mapping = file->f_mapping;
514                 if (!mapping)
515                         mapping = file->f_dentry->d_inode->i_mapping;
516
517                 capabilities = 0;
518                 if (mapping && mapping->backing_dev_info)
519                         capabilities = mapping->backing_dev_info->capabilities;
520
521                 if (!capabilities) {
522                         /* no explicit capabilities set, so assume some
523                          * defaults */
524                         switch (file->f_dentry->d_inode->i_mode & S_IFMT) {
525                         case S_IFREG:
526                         case S_IFBLK:
527                                 capabilities = BDI_CAP_MAP_COPY;
528                                 break;
529
530                         case S_IFCHR:
531                                 capabilities =
532                                         BDI_CAP_MAP_DIRECT |
533                                         BDI_CAP_READ_MAP |
534                                         BDI_CAP_WRITE_MAP;
535                                 break;
536
537                         default:
538                                 return -EINVAL;
539                         }
540                 }
541
542                 /* eliminate any capabilities that we can't support on this
543                  * device */
544                 if (!file->f_op->get_unmapped_area)
545                         capabilities &= ~BDI_CAP_MAP_DIRECT;
546                 if (!file->f_op->read)
547                         capabilities &= ~BDI_CAP_MAP_COPY;
548
549                 if (flags & MAP_SHARED) {
550                         /* do checks for writing, appending and locking */
551                         if ((prot & PROT_WRITE) &&
552                             !(file->f_mode & FMODE_WRITE))
553                                 return -EACCES;
554
555                         if (IS_APPEND(file->f_dentry->d_inode) &&
556                             (file->f_mode & FMODE_WRITE))
557                                 return -EACCES;
558
559                         if (locks_verify_locked(file->f_dentry->d_inode))
560                                 return -EAGAIN;
561
562                         if (!(capabilities & BDI_CAP_MAP_DIRECT))
563                                 return -ENODEV;
564
565                         if (((prot & PROT_READ)  && !(capabilities & BDI_CAP_READ_MAP))  ||
566                             ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
567                             ((prot & PROT_EXEC)  && !(capabilities & BDI_CAP_EXEC_MAP))
568                             ) {
569                                 printk("MAP_SHARED not completely supported on !MMU\n");
570                                 return -EINVAL;
571                         }
572
573                         /* we mustn't privatise shared mappings */
574                         capabilities &= ~BDI_CAP_MAP_COPY;
575                 }
576                 else {
577                         /* we're going to read the file into private memory we
578                          * allocate */
579                         if (!(capabilities & BDI_CAP_MAP_COPY))
580                                 return -ENODEV;
581
582                         /* we don't permit a private writable mapping to be
583                          * shared with the backing device */
584                         if (prot & PROT_WRITE)
585                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
586                 }
587
588                 /* handle executable mappings and implied executable
589                  * mappings */
590                 if (file->f_vfsmnt->mnt_flags & MNT_NOEXEC) {
591                         if (prot & PROT_EXEC)
592                                 return -EPERM;
593                 }
594                 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
595                         /* handle implication of PROT_EXEC by PROT_READ */
596                         if (current->personality & READ_IMPLIES_EXEC) {
597                                 if (capabilities & BDI_CAP_EXEC_MAP)
598                                         prot |= PROT_EXEC;
599                         }
600                 }
601                 else if ((prot & PROT_READ) &&
602                          (prot & PROT_EXEC) &&
603                          !(capabilities & BDI_CAP_EXEC_MAP)
604                          ) {
605                         /* backing file is not executable, try to copy */
606                         capabilities &= ~BDI_CAP_MAP_DIRECT;
607                 }
608         }
609         else {
610                 /* anonymous mappings are always memory backed and can be
611                  * privately mapped
612                  */
613                 capabilities = BDI_CAP_MAP_COPY;
614
615                 /* handle PROT_EXEC implication by PROT_READ */
616                 if ((prot & PROT_READ) &&
617                     (current->personality & READ_IMPLIES_EXEC))
618                         prot |= PROT_EXEC;
619         }
620
621         /* allow the security API to have its say */
622         ret = security_file_mmap(file, reqprot, prot, flags);
623         if (ret < 0)
624                 return ret;
625
626         /* looks okay */
627         *_capabilities = capabilities;
628         return 0;
629 }
630
631 /*
632  * we've determined that we can make the mapping, now translate what we
633  * now know into VMA flags
634  */
635 static unsigned long determine_vm_flags(struct file *file,
636                                         unsigned long prot,
637                                         unsigned long flags,
638                                         unsigned long capabilities)
639 {
640         unsigned long vm_flags;
641
642         vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
643         vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
644         /* vm_flags |= mm->def_flags; */
645
646         if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
647                 /* attempt to share read-only copies of mapped file chunks */
648                 if (file && !(prot & PROT_WRITE))
649                         vm_flags |= VM_MAYSHARE;
650         }
651         else {
652                 /* overlay a shareable mapping on the backing device or inode
653                  * if possible - used for chardevs, ramfs/tmpfs/shmfs and
654                  * romfs/cramfs */
655                 if (flags & MAP_SHARED)
656                         vm_flags |= VM_MAYSHARE | VM_SHARED;
657                 else if ((((vm_flags & capabilities) ^ vm_flags) & BDI_CAP_VMFLAGS) == 0)
658                         vm_flags |= VM_MAYSHARE;
659         }
660
661         /* refuse to let anyone share private mappings with this process if
662          * it's being traced - otherwise breakpoints set in it may interfere
663          * with another untraced process
664          */
665         if ((flags & MAP_PRIVATE) && (current->ptrace & PT_PTRACED))
666                 vm_flags &= ~VM_MAYSHARE;
667
668         return vm_flags;
669 }
670
671 /*
672  * set up a shared mapping on a file
673  */
674 static int do_mmap_shared_file(struct vm_area_struct *vma, unsigned long len)
675 {
676         int ret;
677
678         ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
679         if (ret != -ENOSYS)
680                 return ret;
681
682         /* getting an ENOSYS error indicates that direct mmap isn't
683          * possible (as opposed to tried but failed) so we'll fall
684          * through to making a private copy of the data and mapping
685          * that if we can */
686         return -ENODEV;
687 }
688
689 /*
690  * set up a private mapping or an anonymous shared mapping
691  */
692 static int do_mmap_private(struct vm_area_struct *vma, unsigned long len)
693 {
694         void *base;
695         int ret;
696
697         /* invoke the file's mapping function so that it can keep track of
698          * shared mappings on devices or memory
699          * - VM_MAYSHARE will be set if it may attempt to share
700          */
701         if (vma->vm_file) {
702                 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
703                 if (ret != -ENOSYS) {
704                         /* shouldn't return success if we're not sharing */
705                         BUG_ON(ret == 0 && !(vma->vm_flags & VM_MAYSHARE));
706                         return ret; /* success or a real error */
707                 }
708
709                 /* getting an ENOSYS error indicates that direct mmap isn't
710                  * possible (as opposed to tried but failed) so we'll try to
711                  * make a private copy of the data and map that instead */
712         }
713
714         /* allocate some memory to hold the mapping
715          * - note that this may not return a page-aligned address if the object
716          *   we're allocating is smaller than a page
717          */
718         base = kmalloc(len, GFP_KERNEL|__GFP_COMP);
719         if (!base)
720                 goto enomem;
721
722         vma->vm_start = (unsigned long) base;
723         vma->vm_end = vma->vm_start + len;
724         vma->vm_flags |= VM_MAPPED_COPY;
725
726 #ifdef WARN_ON_SLACK
727         if (len + WARN_ON_SLACK <= kobjsize(result))
728                 printk("Allocation of %lu bytes from process %d has %lu bytes of slack\n",
729                        len, current->pid, kobjsize(result) - len);
730 #endif
731
732         if (vma->vm_file) {
733                 /* read the contents of a file into the copy */
734                 mm_segment_t old_fs;
735                 loff_t fpos;
736
737                 fpos = vma->vm_pgoff;
738                 fpos <<= PAGE_SHIFT;
739
740                 old_fs = get_fs();
741                 set_fs(KERNEL_DS);
742                 ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos);
743                 set_fs(old_fs);
744
745                 if (ret < 0)
746                         goto error_free;
747
748                 /* clear the last little bit */
749                 if (ret < len)
750                         memset(base + ret, 0, len - ret);
751
752         } else {
753                 /* if it's an anonymous mapping, then just clear it */
754                 memset(base, 0, len);
755         }
756
757         return 0;
758
759 error_free:
760         kfree(base);
761         vma->vm_start = 0;
762         return ret;
763
764 enomem:
765         printk("Allocation of length %lu from process %d failed\n",
766                len, current->pid);
767         show_free_areas();
768         return -ENOMEM;
769 }
770
771 /*
772  * handle mapping creation for uClinux
773  */
774 unsigned long do_mmap_pgoff(struct file *file,
775                             unsigned long addr,
776                             unsigned long len,
777                             unsigned long prot,
778                             unsigned long flags,
779                             unsigned long pgoff)
780 {
781         struct vm_list_struct *vml = NULL;
782         struct vm_area_struct *vma = NULL;
783         struct rb_node *rb;
784         unsigned long capabilities, vm_flags;
785         void *result;
786         int ret;
787
788         /* decide whether we should attempt the mapping, and if so what sort of
789          * mapping */
790         ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
791                                     &capabilities);
792         if (ret < 0)
793                 return ret;
794
795         /* we've determined that we can make the mapping, now translate what we
796          * now know into VMA flags */
797         vm_flags = determine_vm_flags(file, prot, flags, capabilities);
798
799         /* we're going to need to record the mapping if it works */
800         vml = kmalloc(sizeof(struct vm_list_struct), GFP_KERNEL);
801         if (!vml)
802                 goto error_getting_vml;
803         memset(vml, 0, sizeof(*vml));
804
805         down_write(&nommu_vma_sem);
806
807         /* if we want to share, we need to check for VMAs created by other
808          * mmap() calls that overlap with our proposed mapping
809          * - we can only share with an exact match on most regular files
810          * - shared mappings on character devices and memory backed files are
811          *   permitted to overlap inexactly as far as we are concerned for in
812          *   these cases, sharing is handled in the driver or filesystem rather
813          *   than here
814          */
815         if (vm_flags & VM_MAYSHARE) {
816                 unsigned long pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
817                 unsigned long vmpglen;
818
819                 for (rb = rb_first(&nommu_vma_tree); rb; rb = rb_next(rb)) {
820                         vma = rb_entry(rb, struct vm_area_struct, vm_rb);
821
822                         if (!(vma->vm_flags & VM_MAYSHARE))
823                                 continue;
824
825                         /* search for overlapping mappings on the same file */
826                         if (vma->vm_file->f_dentry->d_inode != file->f_dentry->d_inode)
827                                 continue;
828
829                         if (vma->vm_pgoff >= pgoff + pglen)
830                                 continue;
831
832                         vmpglen = vma->vm_end - vma->vm_start + PAGE_SIZE - 1;
833                         vmpglen >>= PAGE_SHIFT;
834                         if (pgoff >= vma->vm_pgoff + vmpglen)
835                                 continue;
836
837                         /* handle inexactly overlapping matches between mappings */
838                         if (vma->vm_pgoff != pgoff || vmpglen != pglen) {
839                                 if (!(capabilities & BDI_CAP_MAP_DIRECT))
840                                         goto sharing_violation;
841                                 continue;
842                         }
843
844                         /* we've found a VMA we can share */
845                         atomic_inc(&vma->vm_usage);
846
847                         vml->vma = vma;
848                         result = (void *) vma->vm_start;
849                         goto shared;
850                 }
851
852                 vma = NULL;
853
854                 /* obtain the address at which to make a shared mapping
855                  * - this is the hook for quasi-memory character devices to
856                  *   tell us the location of a shared mapping
857                  */
858                 if (file && file->f_op->get_unmapped_area) {
859                         addr = file->f_op->get_unmapped_area(file, addr, len,
860                                                              pgoff, flags);
861                         if (IS_ERR((void *) addr)) {
862                                 ret = addr;
863                                 if (ret != (unsigned long) -ENOSYS)
864                                         goto error;
865
866                                 /* the driver refused to tell us where to site
867                                  * the mapping so we'll have to attempt to copy
868                                  * it */
869                                 ret = (unsigned long) -ENODEV;
870                                 if (!(capabilities & BDI_CAP_MAP_COPY))
871                                         goto error;
872
873                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
874                         }
875                 }
876         }
877
878         /* we're going to need a VMA struct as well */
879         vma = kmalloc(sizeof(struct vm_area_struct), GFP_KERNEL);
880         if (!vma)
881                 goto error_getting_vma;
882
883         memset(vma, 0, sizeof(*vma));
884         INIT_LIST_HEAD(&vma->anon_vma_node);
885         atomic_set(&vma->vm_usage, 1);
886         if (file)
887                 get_file(file);
888         vma->vm_file    = file;
889         vma->vm_flags   = vm_flags;
890         vma->vm_start   = addr;
891         vma->vm_end     = addr + len;
892         vma->vm_pgoff   = pgoff;
893
894         vml->vma = vma;
895
896         /* set up the mapping */
897         if (file && vma->vm_flags & VM_SHARED)
898                 ret = do_mmap_shared_file(vma, len);
899         else
900                 ret = do_mmap_private(vma, len);
901         if (ret < 0)
902                 goto error;
903
904         /* okay... we have a mapping; now we have to register it */
905         result = (void *) vma->vm_start;
906
907         if (vma->vm_flags & VM_MAPPED_COPY) {
908                 realalloc += kobjsize(result);
909                 askedalloc += len;
910         }
911
912         realalloc += kobjsize(vma);
913         askedalloc += sizeof(*vma);
914
915         current->mm->total_vm += len >> PAGE_SHIFT;
916
917         add_nommu_vma(vma);
918
919  shared:
920         realalloc += kobjsize(vml);
921         askedalloc += sizeof(*vml);
922
923         add_vma_to_mm(current->mm, vml);
924
925         up_write(&nommu_vma_sem);
926
927         if (prot & PROT_EXEC)
928                 flush_icache_range((unsigned long) result,
929                                    (unsigned long) result + len);
930
931 #ifdef DEBUG
932         printk("do_mmap:\n");
933         show_process_blocks();
934 #endif
935
936         return (unsigned long) result;
937
938  error:
939         up_write(&nommu_vma_sem);
940         kfree(vml);
941         if (vma) {
942                 fput(vma->vm_file);
943                 kfree(vma);
944         }
945         return ret;
946
947  sharing_violation:
948         up_write(&nommu_vma_sem);
949         printk("Attempt to share mismatched mappings\n");
950         kfree(vml);
951         return -EINVAL;
952
953  error_getting_vma:
954         up_write(&nommu_vma_sem);
955         kfree(vml);
956         printk("Allocation of vma for %lu byte allocation from process %d failed\n",
957                len, current->pid);
958         show_free_areas();
959         return -ENOMEM;
960
961  error_getting_vml:
962         printk("Allocation of vml for %lu byte allocation from process %d failed\n",
963                len, current->pid);
964         show_free_areas();
965         return -ENOMEM;
966 }
967
968 /*
969  * handle mapping disposal for uClinux
970  */
971 static void put_vma(struct vm_area_struct *vma)
972 {
973         if (vma) {
974                 down_write(&nommu_vma_sem);
975
976                 if (atomic_dec_and_test(&vma->vm_usage)) {
977                         delete_nommu_vma(vma);
978
979                         if (vma->vm_ops && vma->vm_ops->close)
980                                 vma->vm_ops->close(vma);
981
982                         /* IO memory and memory shared directly out of the pagecache from
983                          * ramfs/tmpfs mustn't be released here */
984                         if (vma->vm_flags & VM_MAPPED_COPY) {
985                                 realalloc -= kobjsize((void *) vma->vm_start);
986                                 askedalloc -= vma->vm_end - vma->vm_start;
987                                 kfree((void *) vma->vm_start);
988                         }
989
990                         realalloc -= kobjsize(vma);
991                         askedalloc -= sizeof(*vma);
992
993                         if (vma->vm_file)
994                                 fput(vma->vm_file);
995                         kfree(vma);
996                 }
997
998                 up_write(&nommu_vma_sem);
999         }
1000 }
1001
1002 /*
1003  * release a mapping
1004  * - under NOMMU conditions the parameters must match exactly to the mapping to
1005  *   be removed
1006  */
1007 int do_munmap(struct mm_struct *mm, unsigned long addr, size_t len)
1008 {
1009         struct vm_list_struct *vml, **parent;
1010         unsigned long end = addr + len;
1011
1012 #ifdef DEBUG
1013         printk("do_munmap:\n");
1014 #endif
1015
1016         for (parent = &mm->context.vmlist; *parent; parent = &(*parent)->next) {
1017                 if ((*parent)->vma->vm_start > addr)
1018                         break;
1019                 if ((*parent)->vma->vm_start == addr &&
1020                     ((len == 0) || ((*parent)->vma->vm_end == end)))
1021                         goto found;
1022         }
1023
1024         printk("munmap of non-mmaped memory by process %d (%s): %p\n",
1025                current->pid, current->comm, (void *) addr);
1026         return -EINVAL;
1027
1028  found:
1029         vml = *parent;
1030
1031         put_vma(vml->vma);
1032
1033         *parent = vml->next;
1034         realalloc -= kobjsize(vml);
1035         askedalloc -= sizeof(*vml);
1036         kfree(vml);
1037
1038         update_hiwater_vm(mm);
1039         mm->total_vm -= len >> PAGE_SHIFT;
1040
1041 #ifdef DEBUG
1042         show_process_blocks();
1043 #endif
1044
1045         return 0;
1046 }
1047
1048 asmlinkage long sys_munmap(unsigned long addr, size_t len)
1049 {
1050         int ret;
1051         struct mm_struct *mm = current->mm;
1052
1053         down_write(&mm->mmap_sem);
1054         ret = do_munmap(mm, addr, len);
1055         up_write(&mm->mmap_sem);
1056         return ret;
1057 }
1058
1059 /*
1060  * Release all mappings
1061  */
1062 void exit_mmap(struct mm_struct * mm)
1063 {
1064         struct vm_list_struct *tmp;
1065
1066         if (mm) {
1067 #ifdef DEBUG
1068                 printk("Exit_mmap:\n");
1069 #endif
1070
1071                 mm->total_vm = 0;
1072
1073                 while ((tmp = mm->context.vmlist)) {
1074                         mm->context.vmlist = tmp->next;
1075                         put_vma(tmp->vma);
1076
1077                         realalloc -= kobjsize(tmp);
1078                         askedalloc -= sizeof(*tmp);
1079                         kfree(tmp);
1080                 }
1081
1082 #ifdef DEBUG
1083                 show_process_blocks();
1084 #endif
1085         }
1086 }
1087
1088 unsigned long do_brk(unsigned long addr, unsigned long len)
1089 {
1090         return -ENOMEM;
1091 }
1092
1093 /*
1094  * expand (or shrink) an existing mapping, potentially moving it at the same
1095  * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1096  *
1097  * under NOMMU conditions, we only permit changing a mapping's size, and only
1098  * as long as it stays within the hole allocated by the kmalloc() call in
1099  * do_mmap_pgoff() and the block is not shareable
1100  *
1101  * MREMAP_FIXED is not supported under NOMMU conditions
1102  */
1103 unsigned long do_mremap(unsigned long addr,
1104                         unsigned long old_len, unsigned long new_len,
1105                         unsigned long flags, unsigned long new_addr)
1106 {
1107         struct vm_area_struct *vma;
1108
1109         /* insanity checks first */
1110         if (new_len == 0)
1111                 return (unsigned long) -EINVAL;
1112
1113         if (flags & MREMAP_FIXED && new_addr != addr)
1114                 return (unsigned long) -EINVAL;
1115
1116         vma = find_vma_exact(current->mm, addr);
1117         if (!vma)
1118                 return (unsigned long) -EINVAL;
1119
1120         if (vma->vm_end != vma->vm_start + old_len)
1121                 return (unsigned long) -EFAULT;
1122
1123         if (vma->vm_flags & VM_MAYSHARE)
1124                 return (unsigned long) -EPERM;
1125
1126         if (new_len > kobjsize((void *) addr))
1127                 return (unsigned long) -ENOMEM;
1128
1129         /* all checks complete - do it */
1130         vma->vm_end = vma->vm_start + new_len;
1131
1132         askedalloc -= old_len;
1133         askedalloc += new_len;
1134
1135         return vma->vm_start;
1136 }
1137
1138 asmlinkage unsigned long sys_mremap(unsigned long addr,
1139         unsigned long old_len, unsigned long new_len,
1140         unsigned long flags, unsigned long new_addr)
1141 {
1142         unsigned long ret;
1143
1144         down_write(&current->mm->mmap_sem);
1145         ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1146         up_write(&current->mm->mmap_sem);
1147         return ret;
1148 }
1149
1150 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1151                         unsigned int foll_flags)
1152 {
1153         return NULL;
1154 }
1155
1156 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
1157 {
1158         return NULL;
1159 }
1160
1161 int remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
1162                 unsigned long to, unsigned long size, pgprot_t prot)
1163 {
1164         vma->vm_start = vma->vm_pgoff << PAGE_SHIFT;
1165         return 0;
1166 }
1167 EXPORT_SYMBOL(remap_pfn_range);
1168
1169 void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1170 {
1171 }
1172
1173 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1174         unsigned long len, unsigned long pgoff, unsigned long flags)
1175 {
1176         return -ENOMEM;
1177 }
1178
1179 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1180 {
1181 }
1182
1183 void unmap_mapping_range(struct address_space *mapping,
1184                          loff_t const holebegin, loff_t const holelen,
1185                          int even_cows)
1186 {
1187 }
1188 EXPORT_SYMBOL(unmap_mapping_range);
1189
1190 /*
1191  * Check that a process has enough memory to allocate a new virtual
1192  * mapping. 0 means there is enough memory for the allocation to
1193  * succeed and -ENOMEM implies there is not.
1194  *
1195  * We currently support three overcommit policies, which are set via the
1196  * vm.overcommit_memory sysctl.  See Documentation/vm/overcommit-accounting
1197  *
1198  * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1199  * Additional code 2002 Jul 20 by Robert Love.
1200  *
1201  * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1202  *
1203  * Note this is a helper function intended to be used by LSMs which
1204  * wish to use this logic.
1205  */
1206 int __vm_enough_memory(long pages, int cap_sys_admin)
1207 {
1208         unsigned long free, allowed;
1209
1210         vm_acct_memory(pages);
1211
1212         /*
1213          * Sometimes we want to use more memory than we have
1214          */
1215         if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1216                 return 0;
1217
1218         if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1219                 unsigned long n;
1220
1221                 free = global_page_state(NR_FILE_PAGES);
1222                 free += nr_swap_pages;
1223
1224                 /*
1225                  * Any slabs which are created with the
1226                  * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1227                  * which are reclaimable, under pressure.  The dentry
1228                  * cache and most inode caches should fall into this
1229                  */
1230                 free += global_page_state(NR_SLAB_RECLAIMABLE);
1231
1232                 /*
1233                  * Leave the last 3% for root
1234                  */
1235                 if (!cap_sys_admin)
1236                         free -= free / 32;
1237
1238                 if (free > pages)
1239                         return 0;
1240
1241                 /*
1242                  * nr_free_pages() is very expensive on large systems,
1243                  * only call if we're about to fail.
1244                  */
1245                 n = nr_free_pages();
1246
1247                 /*
1248                  * Leave reserved pages. The pages are not for anonymous pages.
1249                  */
1250                 if (n <= totalreserve_pages)
1251                         goto error;
1252                 else
1253                         n -= totalreserve_pages;
1254
1255                 /*
1256                  * Leave the last 3% for root
1257                  */
1258                 if (!cap_sys_admin)
1259                         n -= n / 32;
1260                 free += n;
1261
1262                 if (free > pages)
1263                         return 0;
1264
1265                 goto error;
1266         }
1267
1268         allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1269         /*
1270          * Leave the last 3% for root
1271          */
1272         if (!cap_sys_admin)
1273                 allowed -= allowed / 32;
1274         allowed += total_swap_pages;
1275
1276         /* Don't let a single process grow too big:
1277            leave 3% of the size of this process for other processes */
1278         allowed -= current->mm->total_vm / 32;
1279
1280         /*
1281          * cast `allowed' as a signed long because vm_committed_space
1282          * sometimes has a negative value
1283          */
1284         if (atomic_read(&vm_committed_space) < (long)allowed)
1285                 return 0;
1286 error:
1287         vm_unacct_memory(pages);
1288
1289         return -ENOMEM;
1290 }
1291
1292 int in_gate_area_no_task(unsigned long addr)
1293 {
1294         return 0;
1295 }
1296
1297 struct page *filemap_nopage(struct vm_area_struct *area,
1298                         unsigned long address, int *type)
1299 {
1300         BUG();
1301         return NULL;
1302 }
1303
1304 /*
1305  * Access another process' address space.
1306  * - source/target buffer must be kernel space
1307  */
1308 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
1309 {
1310         struct vm_area_struct *vma;
1311         struct mm_struct *mm;
1312
1313         if (addr + len < addr)
1314                 return 0;
1315
1316         mm = get_task_mm(tsk);
1317         if (!mm)
1318                 return 0;
1319
1320         down_read(&mm->mmap_sem);
1321
1322         /* the access must start within one of the target process's mappings */
1323         vma = find_vma(mm, addr);
1324         if (vma) {
1325                 /* don't overrun this mapping */
1326                 if (addr + len >= vma->vm_end)
1327                         len = vma->vm_end - addr;
1328
1329                 /* only read or write mappings where it is permitted */
1330                 if (write && vma->vm_flags & VM_MAYWRITE)
1331                         len -= copy_to_user((void *) addr, buf, len);
1332                 else if (!write && vma->vm_flags & VM_MAYREAD)
1333                         len -= copy_from_user(buf, (void *) addr, len);
1334                 else
1335                         len = 0;
1336         } else {
1337                 len = 0;
1338         }
1339
1340         up_read(&mm->mmap_sem);
1341         mmput(mm);
1342         return len;
1343 }