[PATCH] nommu: fix bug ip_conntrack does not work on nommu
[linux-2.6.git] / mm / nommu.c
1 /*
2  *  linux/mm/nommu.c
3  *
4  *  Replacement code for mm functions to support CPU's that don't
5  *  have any form of memory management unit (thus no virtual memory).
6  *
7  *  See Documentation/nommu-mmap.txt
8  *
9  *  Copyright (c) 2004-2005 David Howells <dhowells@redhat.com>
10  *  Copyright (c) 2000-2003 David McCullough <davidm@snapgear.com>
11  *  Copyright (c) 2000-2001 D Jeff Dionne <jeff@uClinux.org>
12  *  Copyright (c) 2002      Greg Ungerer <gerg@snapgear.com>
13  */
14
15 #include <linux/mm.h>
16 #include <linux/mman.h>
17 #include <linux/swap.h>
18 #include <linux/file.h>
19 #include <linux/highmem.h>
20 #include <linux/pagemap.h>
21 #include <linux/slab.h>
22 #include <linux/vmalloc.h>
23 #include <linux/ptrace.h>
24 #include <linux/blkdev.h>
25 #include <linux/backing-dev.h>
26 #include <linux/mount.h>
27 #include <linux/personality.h>
28 #include <linux/security.h>
29 #include <linux/syscalls.h>
30
31 #include <asm/uaccess.h>
32 #include <asm/tlb.h>
33 #include <asm/tlbflush.h>
34
35 void *high_memory;
36 struct page *mem_map;
37 unsigned long max_mapnr;
38 unsigned long num_physpages;
39 unsigned long askedalloc, realalloc;
40 atomic_t vm_committed_space = ATOMIC_INIT(0);
41 int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
42 int sysctl_overcommit_ratio = 50; /* default is 50% */
43 int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
44 int heap_stack_gap = 0;
45
46 EXPORT_SYMBOL(mem_map);
47 EXPORT_SYMBOL(__vm_enough_memory);
48 EXPORT_SYMBOL(num_physpages);
49
50 /* list of shareable VMAs */
51 struct rb_root nommu_vma_tree = RB_ROOT;
52 DECLARE_RWSEM(nommu_vma_sem);
53
54 struct vm_operations_struct generic_file_vm_ops = {
55 };
56
57 EXPORT_SYMBOL(vfree);
58 EXPORT_SYMBOL(vmalloc_to_page);
59 EXPORT_SYMBOL(vmalloc_32);
60 EXPORT_SYMBOL(vmap);
61 EXPORT_SYMBOL(vunmap);
62
63 /*
64  * Handle all mappings that got truncated by a "truncate()"
65  * system call.
66  *
67  * NOTE! We have to be ready to update the memory sharing
68  * between the file and the memory map for a potential last
69  * incomplete page.  Ugly, but necessary.
70  */
71 int vmtruncate(struct inode *inode, loff_t offset)
72 {
73         struct address_space *mapping = inode->i_mapping;
74         unsigned long limit;
75
76         if (inode->i_size < offset)
77                 goto do_expand;
78         i_size_write(inode, offset);
79
80         truncate_inode_pages(mapping, offset);
81         goto out_truncate;
82
83 do_expand:
84         limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
85         if (limit != RLIM_INFINITY && offset > limit)
86                 goto out_sig;
87         if (offset > inode->i_sb->s_maxbytes)
88                 goto out;
89         i_size_write(inode, offset);
90
91 out_truncate:
92         if (inode->i_op && inode->i_op->truncate)
93                 inode->i_op->truncate(inode);
94         return 0;
95 out_sig:
96         send_sig(SIGXFSZ, current, 0);
97 out:
98         return -EFBIG;
99 }
100
101 EXPORT_SYMBOL(vmtruncate);
102
103 /*
104  * Return the total memory allocated for this pointer, not
105  * just what the caller asked for.
106  *
107  * Doesn't have to be accurate, i.e. may have races.
108  */
109 unsigned int kobjsize(const void *objp)
110 {
111         struct page *page;
112
113         if (!objp || !((page = virt_to_page(objp))))
114                 return 0;
115
116         if (PageSlab(page))
117                 return ksize(objp);
118
119         BUG_ON(page->index < 0);
120         BUG_ON(page->index >= MAX_ORDER);
121
122         return (PAGE_SIZE << page->index);
123 }
124
125 /*
126  * get a list of pages in an address range belonging to the specified process
127  * and indicate the VMA that covers each page
128  * - this is potentially dodgy as we may end incrementing the page count of a
129  *   slab page or a secondary page from a compound page
130  * - don't permit access to VMAs that don't support it, such as I/O mappings
131  */
132 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
133         unsigned long start, int len, int write, int force,
134         struct page **pages, struct vm_area_struct **vmas)
135 {
136         struct vm_area_struct *vma;
137         unsigned long vm_flags;
138         int i;
139
140         /* calculate required read or write permissions.
141          * - if 'force' is set, we only require the "MAY" flags.
142          */
143         vm_flags  = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
144         vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
145
146         for (i = 0; i < len; i++) {
147                 vma = find_vma(mm, start);
148                 if (!vma)
149                         goto finish_or_fault;
150
151                 /* protect what we can, including chardevs */
152                 if (vma->vm_flags & (VM_IO | VM_PFNMAP) ||
153                     !(vm_flags & vma->vm_flags))
154                         goto finish_or_fault;
155
156                 if (pages) {
157                         pages[i] = virt_to_page(start);
158                         if (pages[i])
159                                 page_cache_get(pages[i]);
160                 }
161                 if (vmas)
162                         vmas[i] = vma;
163                 start += PAGE_SIZE;
164         }
165
166         return i;
167
168 finish_or_fault:
169         return i ? : -EFAULT;
170 }
171
172 EXPORT_SYMBOL(get_user_pages);
173
174 DEFINE_RWLOCK(vmlist_lock);
175 struct vm_struct *vmlist;
176
177 void vfree(void *addr)
178 {
179         kfree(addr);
180 }
181
182 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
183 {
184         /*
185          * kmalloc doesn't like __GFP_HIGHMEM for some reason
186          */
187         return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
188 }
189
190 struct page * vmalloc_to_page(void *addr)
191 {
192         return virt_to_page(addr);
193 }
194
195 unsigned long vmalloc_to_pfn(void *addr)
196 {
197         return page_to_pfn(virt_to_page(addr));
198 }
199
200
201 long vread(char *buf, char *addr, unsigned long count)
202 {
203         memcpy(buf, addr, count);
204         return count;
205 }
206
207 long vwrite(char *buf, char *addr, unsigned long count)
208 {
209         /* Don't allow overflow */
210         if ((unsigned long) addr + count < count)
211                 count = -(unsigned long) addr;
212
213         memcpy(addr, buf, count);
214         return(count);
215 }
216
217 /*
218  *      vmalloc  -  allocate virtually continguos memory
219  *
220  *      @size:          allocation size
221  *
222  *      Allocate enough pages to cover @size from the page level
223  *      allocator and map them into continguos kernel virtual space.
224  *
225  *      For tight control over page level allocator and protection flags
226  *      use __vmalloc() instead.
227  */
228 void *vmalloc(unsigned long size)
229 {
230        return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
231 }
232 EXPORT_SYMBOL(vmalloc);
233
234 void *vmalloc_node(unsigned long size, int node)
235 {
236         return vmalloc(size);
237 }
238 EXPORT_SYMBOL(vmalloc_node);
239
240 /*
241  *      vmalloc_32  -  allocate virtually continguos memory (32bit addressable)
242  *
243  *      @size:          allocation size
244  *
245  *      Allocate enough 32bit PA addressable pages to cover @size from the
246  *      page level allocator and map them into continguos kernel virtual space.
247  */
248 void *vmalloc_32(unsigned long size)
249 {
250         return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
251 }
252
253 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
254 {
255         BUG();
256         return NULL;
257 }
258
259 void vunmap(void *addr)
260 {
261         BUG();
262 }
263
264 /*
265  *  sys_brk() for the most part doesn't need the global kernel
266  *  lock, except when an application is doing something nasty
267  *  like trying to un-brk an area that has already been mapped
268  *  to a regular file.  in this case, the unmapping will need
269  *  to invoke file system routines that need the global lock.
270  */
271 asmlinkage unsigned long sys_brk(unsigned long brk)
272 {
273         struct mm_struct *mm = current->mm;
274
275         if (brk < mm->start_brk || brk > mm->context.end_brk)
276                 return mm->brk;
277
278         if (mm->brk == brk)
279                 return mm->brk;
280
281         /*
282          * Always allow shrinking brk
283          */
284         if (brk <= mm->brk) {
285                 mm->brk = brk;
286                 return brk;
287         }
288
289         /*
290          * Ok, looks good - let it rip.
291          */
292         return mm->brk = brk;
293 }
294
295 #ifdef DEBUG
296 static void show_process_blocks(void)
297 {
298         struct vm_list_struct *vml;
299
300         printk("Process blocks %d:", current->pid);
301
302         for (vml = &current->mm->context.vmlist; vml; vml = vml->next) {
303                 printk(" %p: %p", vml, vml->vma);
304                 if (vml->vma)
305                         printk(" (%d @%lx #%d)",
306                                kobjsize((void *) vml->vma->vm_start),
307                                vml->vma->vm_start,
308                                atomic_read(&vml->vma->vm_usage));
309                 printk(vml->next ? " ->" : ".\n");
310         }
311 }
312 #endif /* DEBUG */
313
314 /*
315  * add a VMA into a process's mm_struct in the appropriate place in the list
316  * - should be called with mm->mmap_sem held writelocked
317  */
318 static void add_vma_to_mm(struct mm_struct *mm, struct vm_list_struct *vml)
319 {
320         struct vm_list_struct **ppv;
321
322         for (ppv = &current->mm->context.vmlist; *ppv; ppv = &(*ppv)->next)
323                 if ((*ppv)->vma->vm_start > vml->vma->vm_start)
324                         break;
325
326         vml->next = *ppv;
327         *ppv = vml;
328 }
329
330 /*
331  * look up the first VMA in which addr resides, NULL if none
332  * - should be called with mm->mmap_sem at least held readlocked
333  */
334 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
335 {
336         struct vm_list_struct *loop, *vml;
337
338         /* search the vm_start ordered list */
339         vml = NULL;
340         for (loop = mm->context.vmlist; loop; loop = loop->next) {
341                 if (loop->vma->vm_start > addr)
342                         break;
343                 vml = loop;
344         }
345
346         if (vml && vml->vma->vm_end > addr)
347                 return vml->vma;
348
349         return NULL;
350 }
351 EXPORT_SYMBOL(find_vma);
352
353 /*
354  * find a VMA
355  * - we don't extend stack VMAs under NOMMU conditions
356  */
357 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
358 {
359         return find_vma(mm, addr);
360 }
361
362 /*
363  * look up the first VMA exactly that exactly matches addr
364  * - should be called with mm->mmap_sem at least held readlocked
365  */
366 static inline struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
367                                                     unsigned long addr)
368 {
369         struct vm_list_struct *vml;
370
371         /* search the vm_start ordered list */
372         for (vml = mm->context.vmlist; vml; vml = vml->next) {
373                 if (vml->vma->vm_start == addr)
374                         return vml->vma;
375                 if (vml->vma->vm_start > addr)
376                         break;
377         }
378
379         return NULL;
380 }
381
382 /*
383  * find a VMA in the global tree
384  */
385 static inline struct vm_area_struct *find_nommu_vma(unsigned long start)
386 {
387         struct vm_area_struct *vma;
388         struct rb_node *n = nommu_vma_tree.rb_node;
389
390         while (n) {
391                 vma = rb_entry(n, struct vm_area_struct, vm_rb);
392
393                 if (start < vma->vm_start)
394                         n = n->rb_left;
395                 else if (start > vma->vm_start)
396                         n = n->rb_right;
397                 else
398                         return vma;
399         }
400
401         return NULL;
402 }
403
404 /*
405  * add a VMA in the global tree
406  */
407 static void add_nommu_vma(struct vm_area_struct *vma)
408 {
409         struct vm_area_struct *pvma;
410         struct address_space *mapping;
411         struct rb_node **p = &nommu_vma_tree.rb_node;
412         struct rb_node *parent = NULL;
413
414         /* add the VMA to the mapping */
415         if (vma->vm_file) {
416                 mapping = vma->vm_file->f_mapping;
417
418                 flush_dcache_mmap_lock(mapping);
419                 vma_prio_tree_insert(vma, &mapping->i_mmap);
420                 flush_dcache_mmap_unlock(mapping);
421         }
422
423         /* add the VMA to the master list */
424         while (*p) {
425                 parent = *p;
426                 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
427
428                 if (vma->vm_start < pvma->vm_start) {
429                         p = &(*p)->rb_left;
430                 }
431                 else if (vma->vm_start > pvma->vm_start) {
432                         p = &(*p)->rb_right;
433                 }
434                 else {
435                         /* mappings are at the same address - this can only
436                          * happen for shared-mem chardevs and shared file
437                          * mappings backed by ramfs/tmpfs */
438                         BUG_ON(!(pvma->vm_flags & VM_SHARED));
439
440                         if (vma < pvma)
441                                 p = &(*p)->rb_left;
442                         else if (vma > pvma)
443                                 p = &(*p)->rb_right;
444                         else
445                                 BUG();
446                 }
447         }
448
449         rb_link_node(&vma->vm_rb, parent, p);
450         rb_insert_color(&vma->vm_rb, &nommu_vma_tree);
451 }
452
453 /*
454  * delete a VMA from the global list
455  */
456 static void delete_nommu_vma(struct vm_area_struct *vma)
457 {
458         struct address_space *mapping;
459
460         /* remove the VMA from the mapping */
461         if (vma->vm_file) {
462                 mapping = vma->vm_file->f_mapping;
463
464                 flush_dcache_mmap_lock(mapping);
465                 vma_prio_tree_remove(vma, &mapping->i_mmap);
466                 flush_dcache_mmap_unlock(mapping);
467         }
468
469         /* remove from the master list */
470         rb_erase(&vma->vm_rb, &nommu_vma_tree);
471 }
472
473 /*
474  * determine whether a mapping should be permitted and, if so, what sort of
475  * mapping we're capable of supporting
476  */
477 static int validate_mmap_request(struct file *file,
478                                  unsigned long addr,
479                                  unsigned long len,
480                                  unsigned long prot,
481                                  unsigned long flags,
482                                  unsigned long pgoff,
483                                  unsigned long *_capabilities)
484 {
485         unsigned long capabilities;
486         unsigned long reqprot = prot;
487         int ret;
488
489         /* do the simple checks first */
490         if (flags & MAP_FIXED || addr) {
491                 printk(KERN_DEBUG
492                        "%d: Can't do fixed-address/overlay mmap of RAM\n",
493                        current->pid);
494                 return -EINVAL;
495         }
496
497         if ((flags & MAP_TYPE) != MAP_PRIVATE &&
498             (flags & MAP_TYPE) != MAP_SHARED)
499                 return -EINVAL;
500
501         if (!len)
502                 return -EINVAL;
503
504         /* Careful about overflows.. */
505         len = PAGE_ALIGN(len);
506         if (!len || len > TASK_SIZE)
507                 return -ENOMEM;
508
509         /* offset overflow? */
510         if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
511                 return -EOVERFLOW;
512
513         if (file) {
514                 /* validate file mapping requests */
515                 struct address_space *mapping;
516
517                 /* files must support mmap */
518                 if (!file->f_op || !file->f_op->mmap)
519                         return -ENODEV;
520
521                 /* work out if what we've got could possibly be shared
522                  * - we support chardevs that provide their own "memory"
523                  * - we support files/blockdevs that are memory backed
524                  */
525                 mapping = file->f_mapping;
526                 if (!mapping)
527                         mapping = file->f_path.dentry->d_inode->i_mapping;
528
529                 capabilities = 0;
530                 if (mapping && mapping->backing_dev_info)
531                         capabilities = mapping->backing_dev_info->capabilities;
532
533                 if (!capabilities) {
534                         /* no explicit capabilities set, so assume some
535                          * defaults */
536                         switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) {
537                         case S_IFREG:
538                         case S_IFBLK:
539                                 capabilities = BDI_CAP_MAP_COPY;
540                                 break;
541
542                         case S_IFCHR:
543                                 capabilities =
544                                         BDI_CAP_MAP_DIRECT |
545                                         BDI_CAP_READ_MAP |
546                                         BDI_CAP_WRITE_MAP;
547                                 break;
548
549                         default:
550                                 return -EINVAL;
551                         }
552                 }
553
554                 /* eliminate any capabilities that we can't support on this
555                  * device */
556                 if (!file->f_op->get_unmapped_area)
557                         capabilities &= ~BDI_CAP_MAP_DIRECT;
558                 if (!file->f_op->read)
559                         capabilities &= ~BDI_CAP_MAP_COPY;
560
561                 if (flags & MAP_SHARED) {
562                         /* do checks for writing, appending and locking */
563                         if ((prot & PROT_WRITE) &&
564                             !(file->f_mode & FMODE_WRITE))
565                                 return -EACCES;
566
567                         if (IS_APPEND(file->f_path.dentry->d_inode) &&
568                             (file->f_mode & FMODE_WRITE))
569                                 return -EACCES;
570
571                         if (locks_verify_locked(file->f_path.dentry->d_inode))
572                                 return -EAGAIN;
573
574                         if (!(capabilities & BDI_CAP_MAP_DIRECT))
575                                 return -ENODEV;
576
577                         if (((prot & PROT_READ)  && !(capabilities & BDI_CAP_READ_MAP))  ||
578                             ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
579                             ((prot & PROT_EXEC)  && !(capabilities & BDI_CAP_EXEC_MAP))
580                             ) {
581                                 printk("MAP_SHARED not completely supported on !MMU\n");
582                                 return -EINVAL;
583                         }
584
585                         /* we mustn't privatise shared mappings */
586                         capabilities &= ~BDI_CAP_MAP_COPY;
587                 }
588                 else {
589                         /* we're going to read the file into private memory we
590                          * allocate */
591                         if (!(capabilities & BDI_CAP_MAP_COPY))
592                                 return -ENODEV;
593
594                         /* we don't permit a private writable mapping to be
595                          * shared with the backing device */
596                         if (prot & PROT_WRITE)
597                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
598                 }
599
600                 /* handle executable mappings and implied executable
601                  * mappings */
602                 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
603                         if (prot & PROT_EXEC)
604                                 return -EPERM;
605                 }
606                 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
607                         /* handle implication of PROT_EXEC by PROT_READ */
608                         if (current->personality & READ_IMPLIES_EXEC) {
609                                 if (capabilities & BDI_CAP_EXEC_MAP)
610                                         prot |= PROT_EXEC;
611                         }
612                 }
613                 else if ((prot & PROT_READ) &&
614                          (prot & PROT_EXEC) &&
615                          !(capabilities & BDI_CAP_EXEC_MAP)
616                          ) {
617                         /* backing file is not executable, try to copy */
618                         capabilities &= ~BDI_CAP_MAP_DIRECT;
619                 }
620         }
621         else {
622                 /* anonymous mappings are always memory backed and can be
623                  * privately mapped
624                  */
625                 capabilities = BDI_CAP_MAP_COPY;
626
627                 /* handle PROT_EXEC implication by PROT_READ */
628                 if ((prot & PROT_READ) &&
629                     (current->personality & READ_IMPLIES_EXEC))
630                         prot |= PROT_EXEC;
631         }
632
633         /* allow the security API to have its say */
634         ret = security_file_mmap(file, reqprot, prot, flags);
635         if (ret < 0)
636                 return ret;
637
638         /* looks okay */
639         *_capabilities = capabilities;
640         return 0;
641 }
642
643 /*
644  * we've determined that we can make the mapping, now translate what we
645  * now know into VMA flags
646  */
647 static unsigned long determine_vm_flags(struct file *file,
648                                         unsigned long prot,
649                                         unsigned long flags,
650                                         unsigned long capabilities)
651 {
652         unsigned long vm_flags;
653
654         vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
655         vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
656         /* vm_flags |= mm->def_flags; */
657
658         if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
659                 /* attempt to share read-only copies of mapped file chunks */
660                 if (file && !(prot & PROT_WRITE))
661                         vm_flags |= VM_MAYSHARE;
662         }
663         else {
664                 /* overlay a shareable mapping on the backing device or inode
665                  * if possible - used for chardevs, ramfs/tmpfs/shmfs and
666                  * romfs/cramfs */
667                 if (flags & MAP_SHARED)
668                         vm_flags |= VM_MAYSHARE | VM_SHARED;
669                 else if ((((vm_flags & capabilities) ^ vm_flags) & BDI_CAP_VMFLAGS) == 0)
670                         vm_flags |= VM_MAYSHARE;
671         }
672
673         /* refuse to let anyone share private mappings with this process if
674          * it's being traced - otherwise breakpoints set in it may interfere
675          * with another untraced process
676          */
677         if ((flags & MAP_PRIVATE) && (current->ptrace & PT_PTRACED))
678                 vm_flags &= ~VM_MAYSHARE;
679
680         return vm_flags;
681 }
682
683 /*
684  * set up a shared mapping on a file
685  */
686 static int do_mmap_shared_file(struct vm_area_struct *vma, unsigned long len)
687 {
688         int ret;
689
690         ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
691         if (ret != -ENOSYS)
692                 return ret;
693
694         /* getting an ENOSYS error indicates that direct mmap isn't
695          * possible (as opposed to tried but failed) so we'll fall
696          * through to making a private copy of the data and mapping
697          * that if we can */
698         return -ENODEV;
699 }
700
701 /*
702  * set up a private mapping or an anonymous shared mapping
703  */
704 static int do_mmap_private(struct vm_area_struct *vma, unsigned long len)
705 {
706         void *base;
707         int ret;
708
709         /* invoke the file's mapping function so that it can keep track of
710          * shared mappings on devices or memory
711          * - VM_MAYSHARE will be set if it may attempt to share
712          */
713         if (vma->vm_file) {
714                 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
715                 if (ret != -ENOSYS) {
716                         /* shouldn't return success if we're not sharing */
717                         BUG_ON(ret == 0 && !(vma->vm_flags & VM_MAYSHARE));
718                         return ret; /* success or a real error */
719                 }
720
721                 /* getting an ENOSYS error indicates that direct mmap isn't
722                  * possible (as opposed to tried but failed) so we'll try to
723                  * make a private copy of the data and map that instead */
724         }
725
726         /* allocate some memory to hold the mapping
727          * - note that this may not return a page-aligned address if the object
728          *   we're allocating is smaller than a page
729          */
730         base = kmalloc(len, GFP_KERNEL|__GFP_COMP);
731         if (!base)
732                 goto enomem;
733
734         vma->vm_start = (unsigned long) base;
735         vma->vm_end = vma->vm_start + len;
736         vma->vm_flags |= VM_MAPPED_COPY;
737
738 #ifdef WARN_ON_SLACK
739         if (len + WARN_ON_SLACK <= kobjsize(result))
740                 printk("Allocation of %lu bytes from process %d has %lu bytes of slack\n",
741                        len, current->pid, kobjsize(result) - len);
742 #endif
743
744         if (vma->vm_file) {
745                 /* read the contents of a file into the copy */
746                 mm_segment_t old_fs;
747                 loff_t fpos;
748
749                 fpos = vma->vm_pgoff;
750                 fpos <<= PAGE_SHIFT;
751
752                 old_fs = get_fs();
753                 set_fs(KERNEL_DS);
754                 ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos);
755                 set_fs(old_fs);
756
757                 if (ret < 0)
758                         goto error_free;
759
760                 /* clear the last little bit */
761                 if (ret < len)
762                         memset(base + ret, 0, len - ret);
763
764         } else {
765                 /* if it's an anonymous mapping, then just clear it */
766                 memset(base, 0, len);
767         }
768
769         return 0;
770
771 error_free:
772         kfree(base);
773         vma->vm_start = 0;
774         return ret;
775
776 enomem:
777         printk("Allocation of length %lu from process %d failed\n",
778                len, current->pid);
779         show_free_areas();
780         return -ENOMEM;
781 }
782
783 /*
784  * handle mapping creation for uClinux
785  */
786 unsigned long do_mmap_pgoff(struct file *file,
787                             unsigned long addr,
788                             unsigned long len,
789                             unsigned long prot,
790                             unsigned long flags,
791                             unsigned long pgoff)
792 {
793         struct vm_list_struct *vml = NULL;
794         struct vm_area_struct *vma = NULL;
795         struct rb_node *rb;
796         unsigned long capabilities, vm_flags;
797         void *result;
798         int ret;
799
800         /* decide whether we should attempt the mapping, and if so what sort of
801          * mapping */
802         ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
803                                     &capabilities);
804         if (ret < 0)
805                 return ret;
806
807         /* we've determined that we can make the mapping, now translate what we
808          * now know into VMA flags */
809         vm_flags = determine_vm_flags(file, prot, flags, capabilities);
810
811         /* we're going to need to record the mapping if it works */
812         vml = kzalloc(sizeof(struct vm_list_struct), GFP_KERNEL);
813         if (!vml)
814                 goto error_getting_vml;
815
816         down_write(&nommu_vma_sem);
817
818         /* if we want to share, we need to check for VMAs created by other
819          * mmap() calls that overlap with our proposed mapping
820          * - we can only share with an exact match on most regular files
821          * - shared mappings on character devices and memory backed files are
822          *   permitted to overlap inexactly as far as we are concerned for in
823          *   these cases, sharing is handled in the driver or filesystem rather
824          *   than here
825          */
826         if (vm_flags & VM_MAYSHARE) {
827                 unsigned long pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
828                 unsigned long vmpglen;
829
830                 /* suppress VMA sharing for shared regions */
831                 if (vm_flags & VM_SHARED &&
832                     capabilities & BDI_CAP_MAP_DIRECT)
833                         goto dont_share_VMAs;
834
835                 for (rb = rb_first(&nommu_vma_tree); rb; rb = rb_next(rb)) {
836                         vma = rb_entry(rb, struct vm_area_struct, vm_rb);
837
838                         if (!(vma->vm_flags & VM_MAYSHARE))
839                                 continue;
840
841                         /* search for overlapping mappings on the same file */
842                         if (vma->vm_file->f_path.dentry->d_inode != file->f_path.dentry->d_inode)
843                                 continue;
844
845                         if (vma->vm_pgoff >= pgoff + pglen)
846                                 continue;
847
848                         vmpglen = vma->vm_end - vma->vm_start + PAGE_SIZE - 1;
849                         vmpglen >>= PAGE_SHIFT;
850                         if (pgoff >= vma->vm_pgoff + vmpglen)
851                                 continue;
852
853                         /* handle inexactly overlapping matches between mappings */
854                         if (vma->vm_pgoff != pgoff || vmpglen != pglen) {
855                                 if (!(capabilities & BDI_CAP_MAP_DIRECT))
856                                         goto sharing_violation;
857                                 continue;
858                         }
859
860                         /* we've found a VMA we can share */
861                         atomic_inc(&vma->vm_usage);
862
863                         vml->vma = vma;
864                         result = (void *) vma->vm_start;
865                         goto shared;
866                 }
867
868         dont_share_VMAs:
869                 vma = NULL;
870
871                 /* obtain the address at which to make a shared mapping
872                  * - this is the hook for quasi-memory character devices to
873                  *   tell us the location of a shared mapping
874                  */
875                 if (file && file->f_op->get_unmapped_area) {
876                         addr = file->f_op->get_unmapped_area(file, addr, len,
877                                                              pgoff, flags);
878                         if (IS_ERR((void *) addr)) {
879                                 ret = addr;
880                                 if (ret != (unsigned long) -ENOSYS)
881                                         goto error;
882
883                                 /* the driver refused to tell us where to site
884                                  * the mapping so we'll have to attempt to copy
885                                  * it */
886                                 ret = (unsigned long) -ENODEV;
887                                 if (!(capabilities & BDI_CAP_MAP_COPY))
888                                         goto error;
889
890                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
891                         }
892                 }
893         }
894
895         /* we're going to need a VMA struct as well */
896         vma = kzalloc(sizeof(struct vm_area_struct), GFP_KERNEL);
897         if (!vma)
898                 goto error_getting_vma;
899
900         INIT_LIST_HEAD(&vma->anon_vma_node);
901         atomic_set(&vma->vm_usage, 1);
902         if (file)
903                 get_file(file);
904         vma->vm_file    = file;
905         vma->vm_flags   = vm_flags;
906         vma->vm_start   = addr;
907         vma->vm_end     = addr + len;
908         vma->vm_pgoff   = pgoff;
909
910         vml->vma = vma;
911
912         /* set up the mapping */
913         if (file && vma->vm_flags & VM_SHARED)
914                 ret = do_mmap_shared_file(vma, len);
915         else
916                 ret = do_mmap_private(vma, len);
917         if (ret < 0)
918                 goto error;
919
920         /* okay... we have a mapping; now we have to register it */
921         result = (void *) vma->vm_start;
922
923         if (vma->vm_flags & VM_MAPPED_COPY) {
924                 realalloc += kobjsize(result);
925                 askedalloc += len;
926         }
927
928         realalloc += kobjsize(vma);
929         askedalloc += sizeof(*vma);
930
931         current->mm->total_vm += len >> PAGE_SHIFT;
932
933         add_nommu_vma(vma);
934
935  shared:
936         realalloc += kobjsize(vml);
937         askedalloc += sizeof(*vml);
938
939         add_vma_to_mm(current->mm, vml);
940
941         up_write(&nommu_vma_sem);
942
943         if (prot & PROT_EXEC)
944                 flush_icache_range((unsigned long) result,
945                                    (unsigned long) result + len);
946
947 #ifdef DEBUG
948         printk("do_mmap:\n");
949         show_process_blocks();
950 #endif
951
952         return (unsigned long) result;
953
954  error:
955         up_write(&nommu_vma_sem);
956         kfree(vml);
957         if (vma) {
958                 if (vma->vm_file)
959                         fput(vma->vm_file);
960                 kfree(vma);
961         }
962         return ret;
963
964  sharing_violation:
965         up_write(&nommu_vma_sem);
966         printk("Attempt to share mismatched mappings\n");
967         kfree(vml);
968         return -EINVAL;
969
970  error_getting_vma:
971         up_write(&nommu_vma_sem);
972         kfree(vml);
973         printk("Allocation of vma for %lu byte allocation from process %d failed\n",
974                len, current->pid);
975         show_free_areas();
976         return -ENOMEM;
977
978  error_getting_vml:
979         printk("Allocation of vml for %lu byte allocation from process %d failed\n",
980                len, current->pid);
981         show_free_areas();
982         return -ENOMEM;
983 }
984
985 /*
986  * handle mapping disposal for uClinux
987  */
988 static void put_vma(struct vm_area_struct *vma)
989 {
990         if (vma) {
991                 down_write(&nommu_vma_sem);
992
993                 if (atomic_dec_and_test(&vma->vm_usage)) {
994                         delete_nommu_vma(vma);
995
996                         if (vma->vm_ops && vma->vm_ops->close)
997                                 vma->vm_ops->close(vma);
998
999                         /* IO memory and memory shared directly out of the pagecache from
1000                          * ramfs/tmpfs mustn't be released here */
1001                         if (vma->vm_flags & VM_MAPPED_COPY) {
1002                                 realalloc -= kobjsize((void *) vma->vm_start);
1003                                 askedalloc -= vma->vm_end - vma->vm_start;
1004                                 kfree((void *) vma->vm_start);
1005                         }
1006
1007                         realalloc -= kobjsize(vma);
1008                         askedalloc -= sizeof(*vma);
1009
1010                         if (vma->vm_file)
1011                                 fput(vma->vm_file);
1012                         kfree(vma);
1013                 }
1014
1015                 up_write(&nommu_vma_sem);
1016         }
1017 }
1018
1019 /*
1020  * release a mapping
1021  * - under NOMMU conditions the parameters must match exactly to the mapping to
1022  *   be removed
1023  */
1024 int do_munmap(struct mm_struct *mm, unsigned long addr, size_t len)
1025 {
1026         struct vm_list_struct *vml, **parent;
1027         unsigned long end = addr + len;
1028
1029 #ifdef DEBUG
1030         printk("do_munmap:\n");
1031 #endif
1032
1033         for (parent = &mm->context.vmlist; *parent; parent = &(*parent)->next) {
1034                 if ((*parent)->vma->vm_start > addr)
1035                         break;
1036                 if ((*parent)->vma->vm_start == addr &&
1037                     ((len == 0) || ((*parent)->vma->vm_end == end)))
1038                         goto found;
1039         }
1040
1041         printk("munmap of non-mmaped memory by process %d (%s): %p\n",
1042                current->pid, current->comm, (void *) addr);
1043         return -EINVAL;
1044
1045  found:
1046         vml = *parent;
1047
1048         put_vma(vml->vma);
1049
1050         *parent = vml->next;
1051         realalloc -= kobjsize(vml);
1052         askedalloc -= sizeof(*vml);
1053         kfree(vml);
1054
1055         update_hiwater_vm(mm);
1056         mm->total_vm -= len >> PAGE_SHIFT;
1057
1058 #ifdef DEBUG
1059         show_process_blocks();
1060 #endif
1061
1062         return 0;
1063 }
1064
1065 asmlinkage long sys_munmap(unsigned long addr, size_t len)
1066 {
1067         int ret;
1068         struct mm_struct *mm = current->mm;
1069
1070         down_write(&mm->mmap_sem);
1071         ret = do_munmap(mm, addr, len);
1072         up_write(&mm->mmap_sem);
1073         return ret;
1074 }
1075
1076 /*
1077  * Release all mappings
1078  */
1079 void exit_mmap(struct mm_struct * mm)
1080 {
1081         struct vm_list_struct *tmp;
1082
1083         if (mm) {
1084 #ifdef DEBUG
1085                 printk("Exit_mmap:\n");
1086 #endif
1087
1088                 mm->total_vm = 0;
1089
1090                 while ((tmp = mm->context.vmlist)) {
1091                         mm->context.vmlist = tmp->next;
1092                         put_vma(tmp->vma);
1093
1094                         realalloc -= kobjsize(tmp);
1095                         askedalloc -= sizeof(*tmp);
1096                         kfree(tmp);
1097                 }
1098
1099 #ifdef DEBUG
1100                 show_process_blocks();
1101 #endif
1102         }
1103 }
1104
1105 unsigned long do_brk(unsigned long addr, unsigned long len)
1106 {
1107         return -ENOMEM;
1108 }
1109
1110 /*
1111  * expand (or shrink) an existing mapping, potentially moving it at the same
1112  * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1113  *
1114  * under NOMMU conditions, we only permit changing a mapping's size, and only
1115  * as long as it stays within the hole allocated by the kmalloc() call in
1116  * do_mmap_pgoff() and the block is not shareable
1117  *
1118  * MREMAP_FIXED is not supported under NOMMU conditions
1119  */
1120 unsigned long do_mremap(unsigned long addr,
1121                         unsigned long old_len, unsigned long new_len,
1122                         unsigned long flags, unsigned long new_addr)
1123 {
1124         struct vm_area_struct *vma;
1125
1126         /* insanity checks first */
1127         if (new_len == 0)
1128                 return (unsigned long) -EINVAL;
1129
1130         if (flags & MREMAP_FIXED && new_addr != addr)
1131                 return (unsigned long) -EINVAL;
1132
1133         vma = find_vma_exact(current->mm, addr);
1134         if (!vma)
1135                 return (unsigned long) -EINVAL;
1136
1137         if (vma->vm_end != vma->vm_start + old_len)
1138                 return (unsigned long) -EFAULT;
1139
1140         if (vma->vm_flags & VM_MAYSHARE)
1141                 return (unsigned long) -EPERM;
1142
1143         if (new_len > kobjsize((void *) addr))
1144                 return (unsigned long) -ENOMEM;
1145
1146         /* all checks complete - do it */
1147         vma->vm_end = vma->vm_start + new_len;
1148
1149         askedalloc -= old_len;
1150         askedalloc += new_len;
1151
1152         return vma->vm_start;
1153 }
1154
1155 asmlinkage unsigned long sys_mremap(unsigned long addr,
1156         unsigned long old_len, unsigned long new_len,
1157         unsigned long flags, unsigned long new_addr)
1158 {
1159         unsigned long ret;
1160
1161         down_write(&current->mm->mmap_sem);
1162         ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1163         up_write(&current->mm->mmap_sem);
1164         return ret;
1165 }
1166
1167 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1168                         unsigned int foll_flags)
1169 {
1170         return NULL;
1171 }
1172
1173 int remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
1174                 unsigned long to, unsigned long size, pgprot_t prot)
1175 {
1176         vma->vm_start = vma->vm_pgoff << PAGE_SHIFT;
1177         return 0;
1178 }
1179 EXPORT_SYMBOL(remap_pfn_range);
1180
1181 void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1182 {
1183 }
1184
1185 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1186         unsigned long len, unsigned long pgoff, unsigned long flags)
1187 {
1188         return -ENOMEM;
1189 }
1190
1191 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1192 {
1193 }
1194
1195 void unmap_mapping_range(struct address_space *mapping,
1196                          loff_t const holebegin, loff_t const holelen,
1197                          int even_cows)
1198 {
1199 }
1200 EXPORT_SYMBOL(unmap_mapping_range);
1201
1202 /*
1203  * ask for an unmapped area at which to create a mapping on a file
1204  */
1205 unsigned long get_unmapped_area(struct file *file, unsigned long addr,
1206                                 unsigned long len, unsigned long pgoff,
1207                                 unsigned long flags)
1208 {
1209         unsigned long (*get_area)(struct file *, unsigned long, unsigned long,
1210                                   unsigned long, unsigned long);
1211
1212         get_area = current->mm->get_unmapped_area;
1213         if (file && file->f_op && file->f_op->get_unmapped_area)
1214                 get_area = file->f_op->get_unmapped_area;
1215
1216         if (!get_area)
1217                 return -ENOSYS;
1218
1219         return get_area(file, addr, len, pgoff, flags);
1220 }
1221
1222 EXPORT_SYMBOL(get_unmapped_area);
1223
1224 /*
1225  * Check that a process has enough memory to allocate a new virtual
1226  * mapping. 0 means there is enough memory for the allocation to
1227  * succeed and -ENOMEM implies there is not.
1228  *
1229  * We currently support three overcommit policies, which are set via the
1230  * vm.overcommit_memory sysctl.  See Documentation/vm/overcommit-accounting
1231  *
1232  * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1233  * Additional code 2002 Jul 20 by Robert Love.
1234  *
1235  * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1236  *
1237  * Note this is a helper function intended to be used by LSMs which
1238  * wish to use this logic.
1239  */
1240 int __vm_enough_memory(long pages, int cap_sys_admin)
1241 {
1242         unsigned long free, allowed;
1243
1244         vm_acct_memory(pages);
1245
1246         /*
1247          * Sometimes we want to use more memory than we have
1248          */
1249         if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1250                 return 0;
1251
1252         if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1253                 unsigned long n;
1254
1255                 free = global_page_state(NR_FILE_PAGES);
1256                 free += nr_swap_pages;
1257
1258                 /*
1259                  * Any slabs which are created with the
1260                  * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1261                  * which are reclaimable, under pressure.  The dentry
1262                  * cache and most inode caches should fall into this
1263                  */
1264                 free += global_page_state(NR_SLAB_RECLAIMABLE);
1265
1266                 /*
1267                  * Leave the last 3% for root
1268                  */
1269                 if (!cap_sys_admin)
1270                         free -= free / 32;
1271
1272                 if (free > pages)
1273                         return 0;
1274
1275                 /*
1276                  * nr_free_pages() is very expensive on large systems,
1277                  * only call if we're about to fail.
1278                  */
1279                 n = nr_free_pages();
1280
1281                 /*
1282                  * Leave reserved pages. The pages are not for anonymous pages.
1283                  */
1284                 if (n <= totalreserve_pages)
1285                         goto error;
1286                 else
1287                         n -= totalreserve_pages;
1288
1289                 /*
1290                  * Leave the last 3% for root
1291                  */
1292                 if (!cap_sys_admin)
1293                         n -= n / 32;
1294                 free += n;
1295
1296                 if (free > pages)
1297                         return 0;
1298
1299                 goto error;
1300         }
1301
1302         allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1303         /*
1304          * Leave the last 3% for root
1305          */
1306         if (!cap_sys_admin)
1307                 allowed -= allowed / 32;
1308         allowed += total_swap_pages;
1309
1310         /* Don't let a single process grow too big:
1311            leave 3% of the size of this process for other processes */
1312         allowed -= current->mm->total_vm / 32;
1313
1314         /*
1315          * cast `allowed' as a signed long because vm_committed_space
1316          * sometimes has a negative value
1317          */
1318         if (atomic_read(&vm_committed_space) < (long)allowed)
1319                 return 0;
1320 error:
1321         vm_unacct_memory(pages);
1322
1323         return -ENOMEM;
1324 }
1325
1326 int in_gate_area_no_task(unsigned long addr)
1327 {
1328         return 0;
1329 }
1330
1331 struct page *filemap_nopage(struct vm_area_struct *area,
1332                         unsigned long address, int *type)
1333 {
1334         BUG();
1335         return NULL;
1336 }
1337
1338 /*
1339  * Access another process' address space.
1340  * - source/target buffer must be kernel space
1341  */
1342 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
1343 {
1344         struct vm_area_struct *vma;
1345         struct mm_struct *mm;
1346
1347         if (addr + len < addr)
1348                 return 0;
1349
1350         mm = get_task_mm(tsk);
1351         if (!mm)
1352                 return 0;
1353
1354         down_read(&mm->mmap_sem);
1355
1356         /* the access must start within one of the target process's mappings */
1357         vma = find_vma(mm, addr);
1358         if (vma) {
1359                 /* don't overrun this mapping */
1360                 if (addr + len >= vma->vm_end)
1361                         len = vma->vm_end - addr;
1362
1363                 /* only read or write mappings where it is permitted */
1364                 if (write && vma->vm_flags & VM_MAYWRITE)
1365                         len -= copy_to_user((void *) addr, buf, len);
1366                 else if (!write && vma->vm_flags & VM_MAYREAD)
1367                         len -= copy_from_user(buf, (void *) addr, len);
1368                 else
1369                         len = 0;
1370         } else {
1371                 len = 0;
1372         }
1373
1374         up_read(&mm->mmap_sem);
1375         mmput(mm);
1376         return len;
1377 }