mm/nommu.c: return 0 from kobjsize with invalid objects
[linux-2.6.git] / mm / nommu.c
1 /*
2  *  linux/mm/nommu.c
3  *
4  *  Replacement code for mm functions to support CPU's that don't
5  *  have any form of memory management unit (thus no virtual memory).
6  *
7  *  See Documentation/nommu-mmap.txt
8  *
9  *  Copyright (c) 2004-2005 David Howells <dhowells@redhat.com>
10  *  Copyright (c) 2000-2003 David McCullough <davidm@snapgear.com>
11  *  Copyright (c) 2000-2001 D Jeff Dionne <jeff@uClinux.org>
12  *  Copyright (c) 2002      Greg Ungerer <gerg@snapgear.com>
13  *  Copyright (c) 2007      Paul Mundt <lethal@linux-sh.org>
14  */
15
16 #include <linux/module.h>
17 #include <linux/mm.h>
18 #include <linux/mman.h>
19 #include <linux/swap.h>
20 #include <linux/file.h>
21 #include <linux/highmem.h>
22 #include <linux/pagemap.h>
23 #include <linux/slab.h>
24 #include <linux/vmalloc.h>
25 #include <linux/ptrace.h>
26 #include <linux/blkdev.h>
27 #include <linux/backing-dev.h>
28 #include <linux/mount.h>
29 #include <linux/personality.h>
30 #include <linux/security.h>
31 #include <linux/syscalls.h>
32
33 #include <asm/uaccess.h>
34 #include <asm/tlb.h>
35 #include <asm/tlbflush.h>
36
37 void *high_memory;
38 struct page *mem_map;
39 unsigned long max_mapnr;
40 unsigned long num_physpages;
41 unsigned long askedalloc, realalloc;
42 atomic_t vm_committed_space = ATOMIC_INIT(0);
43 int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
44 int sysctl_overcommit_ratio = 50; /* default is 50% */
45 int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
46 int heap_stack_gap = 0;
47
48 EXPORT_SYMBOL(mem_map);
49 EXPORT_SYMBOL(num_physpages);
50
51 /* list of shareable VMAs */
52 struct rb_root nommu_vma_tree = RB_ROOT;
53 DECLARE_RWSEM(nommu_vma_sem);
54
55 struct vm_operations_struct generic_file_vm_ops = {
56 };
57
58 /*
59  * Handle all mappings that got truncated by a "truncate()"
60  * system call.
61  *
62  * NOTE! We have to be ready to update the memory sharing
63  * between the file and the memory map for a potential last
64  * incomplete page.  Ugly, but necessary.
65  */
66 int vmtruncate(struct inode *inode, loff_t offset)
67 {
68         struct address_space *mapping = inode->i_mapping;
69         unsigned long limit;
70
71         if (inode->i_size < offset)
72                 goto do_expand;
73         i_size_write(inode, offset);
74
75         truncate_inode_pages(mapping, offset);
76         goto out_truncate;
77
78 do_expand:
79         limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
80         if (limit != RLIM_INFINITY && offset > limit)
81                 goto out_sig;
82         if (offset > inode->i_sb->s_maxbytes)
83                 goto out;
84         i_size_write(inode, offset);
85
86 out_truncate:
87         if (inode->i_op && inode->i_op->truncate)
88                 inode->i_op->truncate(inode);
89         return 0;
90 out_sig:
91         send_sig(SIGXFSZ, current, 0);
92 out:
93         return -EFBIG;
94 }
95
96 EXPORT_SYMBOL(vmtruncate);
97
98 /*
99  * Return the total memory allocated for this pointer, not
100  * just what the caller asked for.
101  *
102  * Doesn't have to be accurate, i.e. may have races.
103  */
104 unsigned int kobjsize(const void *objp)
105 {
106         struct page *page;
107
108         /*
109          * If the object we have should not have ksize performed on it,
110          * return size of 0
111          */
112         if (!objp || (unsigned long)objp >= memory_end || !((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 EXPORT_SYMBOL(get_user_pages);
171
172 DEFINE_RWLOCK(vmlist_lock);
173 struct vm_struct *vmlist;
174
175 void vfree(const void *addr)
176 {
177         kfree(addr);
178 }
179 EXPORT_SYMBOL(vfree);
180
181 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
182 {
183         /*
184          *  You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
185          * returns only a logical address.
186          */
187         return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
188 }
189 EXPORT_SYMBOL(__vmalloc);
190
191 void *vmalloc_user(unsigned long size)
192 {
193         void *ret;
194
195         ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
196                         PAGE_KERNEL);
197         if (ret) {
198                 struct vm_area_struct *vma;
199
200                 down_write(&current->mm->mmap_sem);
201                 vma = find_vma(current->mm, (unsigned long)ret);
202                 if (vma)
203                         vma->vm_flags |= VM_USERMAP;
204                 up_write(&current->mm->mmap_sem);
205         }
206
207         return ret;
208 }
209 EXPORT_SYMBOL(vmalloc_user);
210
211 struct page *vmalloc_to_page(const void *addr)
212 {
213         return virt_to_page(addr);
214 }
215 EXPORT_SYMBOL(vmalloc_to_page);
216
217 unsigned long vmalloc_to_pfn(const void *addr)
218 {
219         return page_to_pfn(virt_to_page(addr));
220 }
221 EXPORT_SYMBOL(vmalloc_to_pfn);
222
223 long vread(char *buf, char *addr, unsigned long count)
224 {
225         memcpy(buf, addr, count);
226         return count;
227 }
228
229 long vwrite(char *buf, char *addr, unsigned long count)
230 {
231         /* Don't allow overflow */
232         if ((unsigned long) addr + count < count)
233                 count = -(unsigned long) addr;
234
235         memcpy(addr, buf, count);
236         return(count);
237 }
238
239 /*
240  *      vmalloc  -  allocate virtually continguos memory
241  *
242  *      @size:          allocation size
243  *
244  *      Allocate enough pages to cover @size from the page level
245  *      allocator and map them into continguos kernel virtual space.
246  *
247  *      For tight control over page level allocator and protection flags
248  *      use __vmalloc() instead.
249  */
250 void *vmalloc(unsigned long size)
251 {
252        return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
253 }
254 EXPORT_SYMBOL(vmalloc);
255
256 void *vmalloc_node(unsigned long size, int node)
257 {
258         return vmalloc(size);
259 }
260 EXPORT_SYMBOL(vmalloc_node);
261
262 /**
263  * vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
264  *      @size:          allocation size
265  *
266  *      Allocate enough 32bit PA addressable pages to cover @size from the
267  *      page level allocator and map them into continguos kernel virtual space.
268  */
269 void *vmalloc_32(unsigned long size)
270 {
271         return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
272 }
273 EXPORT_SYMBOL(vmalloc_32);
274
275 /**
276  * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
277  *      @size:          allocation size
278  *
279  * The resulting memory area is 32bit addressable and zeroed so it can be
280  * mapped to userspace without leaking data.
281  *
282  * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
283  * remap_vmalloc_range() are permissible.
284  */
285 void *vmalloc_32_user(unsigned long size)
286 {
287         /*
288          * We'll have to sort out the ZONE_DMA bits for 64-bit,
289          * but for now this can simply use vmalloc_user() directly.
290          */
291         return vmalloc_user(size);
292 }
293 EXPORT_SYMBOL(vmalloc_32_user);
294
295 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
296 {
297         BUG();
298         return NULL;
299 }
300 EXPORT_SYMBOL(vmap);
301
302 void vunmap(const void *addr)
303 {
304         BUG();
305 }
306 EXPORT_SYMBOL(vunmap);
307
308 /*
309  * Implement a stub for vmalloc_sync_all() if the architecture chose not to
310  * have one.
311  */
312 void  __attribute__((weak)) vmalloc_sync_all(void)
313 {
314 }
315
316 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
317                    struct page *page)
318 {
319         return -EINVAL;
320 }
321 EXPORT_SYMBOL(vm_insert_page);
322
323 /*
324  *  sys_brk() for the most part doesn't need the global kernel
325  *  lock, except when an application is doing something nasty
326  *  like trying to un-brk an area that has already been mapped
327  *  to a regular file.  in this case, the unmapping will need
328  *  to invoke file system routines that need the global lock.
329  */
330 asmlinkage unsigned long sys_brk(unsigned long brk)
331 {
332         struct mm_struct *mm = current->mm;
333
334         if (brk < mm->start_brk || brk > mm->context.end_brk)
335                 return mm->brk;
336
337         if (mm->brk == brk)
338                 return mm->brk;
339
340         /*
341          * Always allow shrinking brk
342          */
343         if (brk <= mm->brk) {
344                 mm->brk = brk;
345                 return brk;
346         }
347
348         /*
349          * Ok, looks good - let it rip.
350          */
351         return mm->brk = brk;
352 }
353
354 #ifdef DEBUG
355 static void show_process_blocks(void)
356 {
357         struct vm_list_struct *vml;
358
359         printk("Process blocks %d:", current->pid);
360
361         for (vml = &current->mm->context.vmlist; vml; vml = vml->next) {
362                 printk(" %p: %p", vml, vml->vma);
363                 if (vml->vma)
364                         printk(" (%d @%lx #%d)",
365                                kobjsize((void *) vml->vma->vm_start),
366                                vml->vma->vm_start,
367                                atomic_read(&vml->vma->vm_usage));
368                 printk(vml->next ? " ->" : ".\n");
369         }
370 }
371 #endif /* DEBUG */
372
373 /*
374  * add a VMA into a process's mm_struct in the appropriate place in the list
375  * - should be called with mm->mmap_sem held writelocked
376  */
377 static void add_vma_to_mm(struct mm_struct *mm, struct vm_list_struct *vml)
378 {
379         struct vm_list_struct **ppv;
380
381         for (ppv = &current->mm->context.vmlist; *ppv; ppv = &(*ppv)->next)
382                 if ((*ppv)->vma->vm_start > vml->vma->vm_start)
383                         break;
384
385         vml->next = *ppv;
386         *ppv = vml;
387 }
388
389 /*
390  * look up the first VMA in which addr resides, NULL if none
391  * - should be called with mm->mmap_sem at least held readlocked
392  */
393 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
394 {
395         struct vm_list_struct *loop, *vml;
396
397         /* search the vm_start ordered list */
398         vml = NULL;
399         for (loop = mm->context.vmlist; loop; loop = loop->next) {
400                 if (loop->vma->vm_start > addr)
401                         break;
402                 vml = loop;
403         }
404
405         if (vml && vml->vma->vm_end > addr)
406                 return vml->vma;
407
408         return NULL;
409 }
410 EXPORT_SYMBOL(find_vma);
411
412 /*
413  * find a VMA
414  * - we don't extend stack VMAs under NOMMU conditions
415  */
416 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
417 {
418         return find_vma(mm, addr);
419 }
420
421 int expand_stack(struct vm_area_struct *vma, unsigned long address)
422 {
423         return -ENOMEM;
424 }
425
426 /*
427  * look up the first VMA exactly that exactly matches addr
428  * - should be called with mm->mmap_sem at least held readlocked
429  */
430 static inline struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
431                                                     unsigned long addr)
432 {
433         struct vm_list_struct *vml;
434
435         /* search the vm_start ordered list */
436         for (vml = mm->context.vmlist; vml; vml = vml->next) {
437                 if (vml->vma->vm_start == addr)
438                         return vml->vma;
439                 if (vml->vma->vm_start > addr)
440                         break;
441         }
442
443         return NULL;
444 }
445
446 /*
447  * find a VMA in the global tree
448  */
449 static inline struct vm_area_struct *find_nommu_vma(unsigned long start)
450 {
451         struct vm_area_struct *vma;
452         struct rb_node *n = nommu_vma_tree.rb_node;
453
454         while (n) {
455                 vma = rb_entry(n, struct vm_area_struct, vm_rb);
456
457                 if (start < vma->vm_start)
458                         n = n->rb_left;
459                 else if (start > vma->vm_start)
460                         n = n->rb_right;
461                 else
462                         return vma;
463         }
464
465         return NULL;
466 }
467
468 /*
469  * add a VMA in the global tree
470  */
471 static void add_nommu_vma(struct vm_area_struct *vma)
472 {
473         struct vm_area_struct *pvma;
474         struct address_space *mapping;
475         struct rb_node **p = &nommu_vma_tree.rb_node;
476         struct rb_node *parent = NULL;
477
478         /* add the VMA to the mapping */
479         if (vma->vm_file) {
480                 mapping = vma->vm_file->f_mapping;
481
482                 flush_dcache_mmap_lock(mapping);
483                 vma_prio_tree_insert(vma, &mapping->i_mmap);
484                 flush_dcache_mmap_unlock(mapping);
485         }
486
487         /* add the VMA to the master list */
488         while (*p) {
489                 parent = *p;
490                 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
491
492                 if (vma->vm_start < pvma->vm_start) {
493                         p = &(*p)->rb_left;
494                 }
495                 else if (vma->vm_start > pvma->vm_start) {
496                         p = &(*p)->rb_right;
497                 }
498                 else {
499                         /* mappings are at the same address - this can only
500                          * happen for shared-mem chardevs and shared file
501                          * mappings backed by ramfs/tmpfs */
502                         BUG_ON(!(pvma->vm_flags & VM_SHARED));
503
504                         if (vma < pvma)
505                                 p = &(*p)->rb_left;
506                         else if (vma > pvma)
507                                 p = &(*p)->rb_right;
508                         else
509                                 BUG();
510                 }
511         }
512
513         rb_link_node(&vma->vm_rb, parent, p);
514         rb_insert_color(&vma->vm_rb, &nommu_vma_tree);
515 }
516
517 /*
518  * delete a VMA from the global list
519  */
520 static void delete_nommu_vma(struct vm_area_struct *vma)
521 {
522         struct address_space *mapping;
523
524         /* remove the VMA from the mapping */
525         if (vma->vm_file) {
526                 mapping = vma->vm_file->f_mapping;
527
528                 flush_dcache_mmap_lock(mapping);
529                 vma_prio_tree_remove(vma, &mapping->i_mmap);
530                 flush_dcache_mmap_unlock(mapping);
531         }
532
533         /* remove from the master list */
534         rb_erase(&vma->vm_rb, &nommu_vma_tree);
535 }
536
537 /*
538  * determine whether a mapping should be permitted and, if so, what sort of
539  * mapping we're capable of supporting
540  */
541 static int validate_mmap_request(struct file *file,
542                                  unsigned long addr,
543                                  unsigned long len,
544                                  unsigned long prot,
545                                  unsigned long flags,
546                                  unsigned long pgoff,
547                                  unsigned long *_capabilities)
548 {
549         unsigned long capabilities;
550         unsigned long reqprot = prot;
551         int ret;
552
553         /* do the simple checks first */
554         if (flags & MAP_FIXED || addr) {
555                 printk(KERN_DEBUG
556                        "%d: Can't do fixed-address/overlay mmap of RAM\n",
557                        current->pid);
558                 return -EINVAL;
559         }
560
561         if ((flags & MAP_TYPE) != MAP_PRIVATE &&
562             (flags & MAP_TYPE) != MAP_SHARED)
563                 return -EINVAL;
564
565         if (!len)
566                 return -EINVAL;
567
568         /* Careful about overflows.. */
569         len = PAGE_ALIGN(len);
570         if (!len || len > TASK_SIZE)
571                 return -ENOMEM;
572
573         /* offset overflow? */
574         if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
575                 return -EOVERFLOW;
576
577         if (file) {
578                 /* validate file mapping requests */
579                 struct address_space *mapping;
580
581                 /* files must support mmap */
582                 if (!file->f_op || !file->f_op->mmap)
583                         return -ENODEV;
584
585                 /* work out if what we've got could possibly be shared
586                  * - we support chardevs that provide their own "memory"
587                  * - we support files/blockdevs that are memory backed
588                  */
589                 mapping = file->f_mapping;
590                 if (!mapping)
591                         mapping = file->f_path.dentry->d_inode->i_mapping;
592
593                 capabilities = 0;
594                 if (mapping && mapping->backing_dev_info)
595                         capabilities = mapping->backing_dev_info->capabilities;
596
597                 if (!capabilities) {
598                         /* no explicit capabilities set, so assume some
599                          * defaults */
600                         switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) {
601                         case S_IFREG:
602                         case S_IFBLK:
603                                 capabilities = BDI_CAP_MAP_COPY;
604                                 break;
605
606                         case S_IFCHR:
607                                 capabilities =
608                                         BDI_CAP_MAP_DIRECT |
609                                         BDI_CAP_READ_MAP |
610                                         BDI_CAP_WRITE_MAP;
611                                 break;
612
613                         default:
614                                 return -EINVAL;
615                         }
616                 }
617
618                 /* eliminate any capabilities that we can't support on this
619                  * device */
620                 if (!file->f_op->get_unmapped_area)
621                         capabilities &= ~BDI_CAP_MAP_DIRECT;
622                 if (!file->f_op->read)
623                         capabilities &= ~BDI_CAP_MAP_COPY;
624
625                 if (flags & MAP_SHARED) {
626                         /* do checks for writing, appending and locking */
627                         if ((prot & PROT_WRITE) &&
628                             !(file->f_mode & FMODE_WRITE))
629                                 return -EACCES;
630
631                         if (IS_APPEND(file->f_path.dentry->d_inode) &&
632                             (file->f_mode & FMODE_WRITE))
633                                 return -EACCES;
634
635                         if (locks_verify_locked(file->f_path.dentry->d_inode))
636                                 return -EAGAIN;
637
638                         if (!(capabilities & BDI_CAP_MAP_DIRECT))
639                                 return -ENODEV;
640
641                         if (((prot & PROT_READ)  && !(capabilities & BDI_CAP_READ_MAP))  ||
642                             ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
643                             ((prot & PROT_EXEC)  && !(capabilities & BDI_CAP_EXEC_MAP))
644                             ) {
645                                 printk("MAP_SHARED not completely supported on !MMU\n");
646                                 return -EINVAL;
647                         }
648
649                         /* we mustn't privatise shared mappings */
650                         capabilities &= ~BDI_CAP_MAP_COPY;
651                 }
652                 else {
653                         /* we're going to read the file into private memory we
654                          * allocate */
655                         if (!(capabilities & BDI_CAP_MAP_COPY))
656                                 return -ENODEV;
657
658                         /* we don't permit a private writable mapping to be
659                          * shared with the backing device */
660                         if (prot & PROT_WRITE)
661                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
662                 }
663
664                 /* handle executable mappings and implied executable
665                  * mappings */
666                 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
667                         if (prot & PROT_EXEC)
668                                 return -EPERM;
669                 }
670                 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
671                         /* handle implication of PROT_EXEC by PROT_READ */
672                         if (current->personality & READ_IMPLIES_EXEC) {
673                                 if (capabilities & BDI_CAP_EXEC_MAP)
674                                         prot |= PROT_EXEC;
675                         }
676                 }
677                 else if ((prot & PROT_READ) &&
678                          (prot & PROT_EXEC) &&
679                          !(capabilities & BDI_CAP_EXEC_MAP)
680                          ) {
681                         /* backing file is not executable, try to copy */
682                         capabilities &= ~BDI_CAP_MAP_DIRECT;
683                 }
684         }
685         else {
686                 /* anonymous mappings are always memory backed and can be
687                  * privately mapped
688                  */
689                 capabilities = BDI_CAP_MAP_COPY;
690
691                 /* handle PROT_EXEC implication by PROT_READ */
692                 if ((prot & PROT_READ) &&
693                     (current->personality & READ_IMPLIES_EXEC))
694                         prot |= PROT_EXEC;
695         }
696
697         /* allow the security API to have its say */
698         ret = security_file_mmap(file, reqprot, prot, flags, addr, 0);
699         if (ret < 0)
700                 return ret;
701
702         /* looks okay */
703         *_capabilities = capabilities;
704         return 0;
705 }
706
707 /*
708  * we've determined that we can make the mapping, now translate what we
709  * now know into VMA flags
710  */
711 static unsigned long determine_vm_flags(struct file *file,
712                                         unsigned long prot,
713                                         unsigned long flags,
714                                         unsigned long capabilities)
715 {
716         unsigned long vm_flags;
717
718         vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
719         vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
720         /* vm_flags |= mm->def_flags; */
721
722         if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
723                 /* attempt to share read-only copies of mapped file chunks */
724                 if (file && !(prot & PROT_WRITE))
725                         vm_flags |= VM_MAYSHARE;
726         }
727         else {
728                 /* overlay a shareable mapping on the backing device or inode
729                  * if possible - used for chardevs, ramfs/tmpfs/shmfs and
730                  * romfs/cramfs */
731                 if (flags & MAP_SHARED)
732                         vm_flags |= VM_MAYSHARE | VM_SHARED;
733                 else if ((((vm_flags & capabilities) ^ vm_flags) & BDI_CAP_VMFLAGS) == 0)
734                         vm_flags |= VM_MAYSHARE;
735         }
736
737         /* refuse to let anyone share private mappings with this process if
738          * it's being traced - otherwise breakpoints set in it may interfere
739          * with another untraced process
740          */
741         if ((flags & MAP_PRIVATE) && (current->ptrace & PT_PTRACED))
742                 vm_flags &= ~VM_MAYSHARE;
743
744         return vm_flags;
745 }
746
747 /*
748  * set up a shared mapping on a file
749  */
750 static int do_mmap_shared_file(struct vm_area_struct *vma, unsigned long len)
751 {
752         int ret;
753
754         ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
755         if (ret != -ENOSYS)
756                 return ret;
757
758         /* getting an ENOSYS error indicates that direct mmap isn't
759          * possible (as opposed to tried but failed) so we'll fall
760          * through to making a private copy of the data and mapping
761          * that if we can */
762         return -ENODEV;
763 }
764
765 /*
766  * set up a private mapping or an anonymous shared mapping
767  */
768 static int do_mmap_private(struct vm_area_struct *vma, unsigned long len)
769 {
770         void *base;
771         int ret;
772
773         /* invoke the file's mapping function so that it can keep track of
774          * shared mappings on devices or memory
775          * - VM_MAYSHARE will be set if it may attempt to share
776          */
777         if (vma->vm_file) {
778                 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
779                 if (ret != -ENOSYS) {
780                         /* shouldn't return success if we're not sharing */
781                         BUG_ON(ret == 0 && !(vma->vm_flags & VM_MAYSHARE));
782                         return ret; /* success or a real error */
783                 }
784
785                 /* getting an ENOSYS error indicates that direct mmap isn't
786                  * possible (as opposed to tried but failed) so we'll try to
787                  * make a private copy of the data and map that instead */
788         }
789
790         /* allocate some memory to hold the mapping
791          * - note that this may not return a page-aligned address if the object
792          *   we're allocating is smaller than a page
793          */
794         base = kmalloc(len, GFP_KERNEL|__GFP_COMP);
795         if (!base)
796                 goto enomem;
797
798         vma->vm_start = (unsigned long) base;
799         vma->vm_end = vma->vm_start + len;
800         vma->vm_flags |= VM_MAPPED_COPY;
801
802 #ifdef WARN_ON_SLACK
803         if (len + WARN_ON_SLACK <= kobjsize(result))
804                 printk("Allocation of %lu bytes from process %d has %lu bytes of slack\n",
805                        len, current->pid, kobjsize(result) - len);
806 #endif
807
808         if (vma->vm_file) {
809                 /* read the contents of a file into the copy */
810                 mm_segment_t old_fs;
811                 loff_t fpos;
812
813                 fpos = vma->vm_pgoff;
814                 fpos <<= PAGE_SHIFT;
815
816                 old_fs = get_fs();
817                 set_fs(KERNEL_DS);
818                 ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos);
819                 set_fs(old_fs);
820
821                 if (ret < 0)
822                         goto error_free;
823
824                 /* clear the last little bit */
825                 if (ret < len)
826                         memset(base + ret, 0, len - ret);
827
828         } else {
829                 /* if it's an anonymous mapping, then just clear it */
830                 memset(base, 0, len);
831         }
832
833         return 0;
834
835 error_free:
836         kfree(base);
837         vma->vm_start = 0;
838         return ret;
839
840 enomem:
841         printk("Allocation of length %lu from process %d failed\n",
842                len, current->pid);
843         show_free_areas();
844         return -ENOMEM;
845 }
846
847 /*
848  * handle mapping creation for uClinux
849  */
850 unsigned long do_mmap_pgoff(struct file *file,
851                             unsigned long addr,
852                             unsigned long len,
853                             unsigned long prot,
854                             unsigned long flags,
855                             unsigned long pgoff)
856 {
857         struct vm_list_struct *vml = NULL;
858         struct vm_area_struct *vma = NULL;
859         struct rb_node *rb;
860         unsigned long capabilities, vm_flags;
861         void *result;
862         int ret;
863
864         if (!(flags & MAP_FIXED))
865                 addr = round_hint_to_min(addr);
866
867         /* decide whether we should attempt the mapping, and if so what sort of
868          * mapping */
869         ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
870                                     &capabilities);
871         if (ret < 0)
872                 return ret;
873
874         /* we've determined that we can make the mapping, now translate what we
875          * now know into VMA flags */
876         vm_flags = determine_vm_flags(file, prot, flags, capabilities);
877
878         /* we're going to need to record the mapping if it works */
879         vml = kzalloc(sizeof(struct vm_list_struct), GFP_KERNEL);
880         if (!vml)
881                 goto error_getting_vml;
882
883         down_write(&nommu_vma_sem);
884
885         /* if we want to share, we need to check for VMAs created by other
886          * mmap() calls that overlap with our proposed mapping
887          * - we can only share with an exact match on most regular files
888          * - shared mappings on character devices and memory backed files are
889          *   permitted to overlap inexactly as far as we are concerned for in
890          *   these cases, sharing is handled in the driver or filesystem rather
891          *   than here
892          */
893         if (vm_flags & VM_MAYSHARE) {
894                 unsigned long pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
895                 unsigned long vmpglen;
896
897                 /* suppress VMA sharing for shared regions */
898                 if (vm_flags & VM_SHARED &&
899                     capabilities & BDI_CAP_MAP_DIRECT)
900                         goto dont_share_VMAs;
901
902                 for (rb = rb_first(&nommu_vma_tree); rb; rb = rb_next(rb)) {
903                         vma = rb_entry(rb, struct vm_area_struct, vm_rb);
904
905                         if (!(vma->vm_flags & VM_MAYSHARE))
906                                 continue;
907
908                         /* search for overlapping mappings on the same file */
909                         if (vma->vm_file->f_path.dentry->d_inode != file->f_path.dentry->d_inode)
910                                 continue;
911
912                         if (vma->vm_pgoff >= pgoff + pglen)
913                                 continue;
914
915                         vmpglen = vma->vm_end - vma->vm_start + PAGE_SIZE - 1;
916                         vmpglen >>= PAGE_SHIFT;
917                         if (pgoff >= vma->vm_pgoff + vmpglen)
918                                 continue;
919
920                         /* handle inexactly overlapping matches between mappings */
921                         if (vma->vm_pgoff != pgoff || vmpglen != pglen) {
922                                 if (!(capabilities & BDI_CAP_MAP_DIRECT))
923                                         goto sharing_violation;
924                                 continue;
925                         }
926
927                         /* we've found a VMA we can share */
928                         atomic_inc(&vma->vm_usage);
929
930                         vml->vma = vma;
931                         result = (void *) vma->vm_start;
932                         goto shared;
933                 }
934
935         dont_share_VMAs:
936                 vma = NULL;
937
938                 /* obtain the address at which to make a shared mapping
939                  * - this is the hook for quasi-memory character devices to
940                  *   tell us the location of a shared mapping
941                  */
942                 if (file && file->f_op->get_unmapped_area) {
943                         addr = file->f_op->get_unmapped_area(file, addr, len,
944                                                              pgoff, flags);
945                         if (IS_ERR((void *) addr)) {
946                                 ret = addr;
947                                 if (ret != (unsigned long) -ENOSYS)
948                                         goto error;
949
950                                 /* the driver refused to tell us where to site
951                                  * the mapping so we'll have to attempt to copy
952                                  * it */
953                                 ret = (unsigned long) -ENODEV;
954                                 if (!(capabilities & BDI_CAP_MAP_COPY))
955                                         goto error;
956
957                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
958                         }
959                 }
960         }
961
962         /* we're going to need a VMA struct as well */
963         vma = kzalloc(sizeof(struct vm_area_struct), GFP_KERNEL);
964         if (!vma)
965                 goto error_getting_vma;
966
967         INIT_LIST_HEAD(&vma->anon_vma_node);
968         atomic_set(&vma->vm_usage, 1);
969         if (file)
970                 get_file(file);
971         vma->vm_file    = file;
972         vma->vm_flags   = vm_flags;
973         vma->vm_start   = addr;
974         vma->vm_end     = addr + len;
975         vma->vm_pgoff   = pgoff;
976
977         vml->vma = vma;
978
979         /* set up the mapping */
980         if (file && vma->vm_flags & VM_SHARED)
981                 ret = do_mmap_shared_file(vma, len);
982         else
983                 ret = do_mmap_private(vma, len);
984         if (ret < 0)
985                 goto error;
986
987         /* okay... we have a mapping; now we have to register it */
988         result = (void *) vma->vm_start;
989
990         if (vma->vm_flags & VM_MAPPED_COPY) {
991                 realalloc += kobjsize(result);
992                 askedalloc += len;
993         }
994
995         realalloc += kobjsize(vma);
996         askedalloc += sizeof(*vma);
997
998         current->mm->total_vm += len >> PAGE_SHIFT;
999
1000         add_nommu_vma(vma);
1001
1002  shared:
1003         realalloc += kobjsize(vml);
1004         askedalloc += sizeof(*vml);
1005
1006         add_vma_to_mm(current->mm, vml);
1007
1008         up_write(&nommu_vma_sem);
1009
1010         if (prot & PROT_EXEC)
1011                 flush_icache_range((unsigned long) result,
1012                                    (unsigned long) result + len);
1013
1014 #ifdef DEBUG
1015         printk("do_mmap:\n");
1016         show_process_blocks();
1017 #endif
1018
1019         return (unsigned long) result;
1020
1021  error:
1022         up_write(&nommu_vma_sem);
1023         kfree(vml);
1024         if (vma) {
1025                 if (vma->vm_file)
1026                         fput(vma->vm_file);
1027                 kfree(vma);
1028         }
1029         return ret;
1030
1031  sharing_violation:
1032         up_write(&nommu_vma_sem);
1033         printk("Attempt to share mismatched mappings\n");
1034         kfree(vml);
1035         return -EINVAL;
1036
1037  error_getting_vma:
1038         up_write(&nommu_vma_sem);
1039         kfree(vml);
1040         printk("Allocation of vma for %lu byte allocation from process %d failed\n",
1041                len, current->pid);
1042         show_free_areas();
1043         return -ENOMEM;
1044
1045  error_getting_vml:
1046         printk("Allocation of vml for %lu byte allocation from process %d failed\n",
1047                len, current->pid);
1048         show_free_areas();
1049         return -ENOMEM;
1050 }
1051 EXPORT_SYMBOL(do_mmap_pgoff);
1052
1053 /*
1054  * handle mapping disposal for uClinux
1055  */
1056 static void put_vma(struct vm_area_struct *vma)
1057 {
1058         if (vma) {
1059                 down_write(&nommu_vma_sem);
1060
1061                 if (atomic_dec_and_test(&vma->vm_usage)) {
1062                         delete_nommu_vma(vma);
1063
1064                         if (vma->vm_ops && vma->vm_ops->close)
1065                                 vma->vm_ops->close(vma);
1066
1067                         /* IO memory and memory shared directly out of the pagecache from
1068                          * ramfs/tmpfs mustn't be released here */
1069                         if (vma->vm_flags & VM_MAPPED_COPY) {
1070                                 realalloc -= kobjsize((void *) vma->vm_start);
1071                                 askedalloc -= vma->vm_end - vma->vm_start;
1072                                 kfree((void *) vma->vm_start);
1073                         }
1074
1075                         realalloc -= kobjsize(vma);
1076                         askedalloc -= sizeof(*vma);
1077
1078                         if (vma->vm_file)
1079                                 fput(vma->vm_file);
1080                         kfree(vma);
1081                 }
1082
1083                 up_write(&nommu_vma_sem);
1084         }
1085 }
1086
1087 /*
1088  * release a mapping
1089  * - under NOMMU conditions the parameters must match exactly to the mapping to
1090  *   be removed
1091  */
1092 int do_munmap(struct mm_struct *mm, unsigned long addr, size_t len)
1093 {
1094         struct vm_list_struct *vml, **parent;
1095         unsigned long end = addr + len;
1096
1097 #ifdef DEBUG
1098         printk("do_munmap:\n");
1099 #endif
1100
1101         for (parent = &mm->context.vmlist; *parent; parent = &(*parent)->next) {
1102                 if ((*parent)->vma->vm_start > addr)
1103                         break;
1104                 if ((*parent)->vma->vm_start == addr &&
1105                     ((len == 0) || ((*parent)->vma->vm_end == end)))
1106                         goto found;
1107         }
1108
1109         printk("munmap of non-mmaped memory by process %d (%s): %p\n",
1110                current->pid, current->comm, (void *) addr);
1111         return -EINVAL;
1112
1113  found:
1114         vml = *parent;
1115
1116         put_vma(vml->vma);
1117
1118         *parent = vml->next;
1119         realalloc -= kobjsize(vml);
1120         askedalloc -= sizeof(*vml);
1121         kfree(vml);
1122
1123         update_hiwater_vm(mm);
1124         mm->total_vm -= len >> PAGE_SHIFT;
1125
1126 #ifdef DEBUG
1127         show_process_blocks();
1128 #endif
1129
1130         return 0;
1131 }
1132 EXPORT_SYMBOL(do_munmap);
1133
1134 asmlinkage long sys_munmap(unsigned long addr, size_t len)
1135 {
1136         int ret;
1137         struct mm_struct *mm = current->mm;
1138
1139         down_write(&mm->mmap_sem);
1140         ret = do_munmap(mm, addr, len);
1141         up_write(&mm->mmap_sem);
1142         return ret;
1143 }
1144
1145 /*
1146  * Release all mappings
1147  */
1148 void exit_mmap(struct mm_struct * mm)
1149 {
1150         struct vm_list_struct *tmp;
1151
1152         if (mm) {
1153 #ifdef DEBUG
1154                 printk("Exit_mmap:\n");
1155 #endif
1156
1157                 mm->total_vm = 0;
1158
1159                 while ((tmp = mm->context.vmlist)) {
1160                         mm->context.vmlist = tmp->next;
1161                         put_vma(tmp->vma);
1162
1163                         realalloc -= kobjsize(tmp);
1164                         askedalloc -= sizeof(*tmp);
1165                         kfree(tmp);
1166                 }
1167
1168 #ifdef DEBUG
1169                 show_process_blocks();
1170 #endif
1171         }
1172 }
1173
1174 unsigned long do_brk(unsigned long addr, unsigned long len)
1175 {
1176         return -ENOMEM;
1177 }
1178
1179 /*
1180  * expand (or shrink) an existing mapping, potentially moving it at the same
1181  * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1182  *
1183  * under NOMMU conditions, we only permit changing a mapping's size, and only
1184  * as long as it stays within the hole allocated by the kmalloc() call in
1185  * do_mmap_pgoff() and the block is not shareable
1186  *
1187  * MREMAP_FIXED is not supported under NOMMU conditions
1188  */
1189 unsigned long do_mremap(unsigned long addr,
1190                         unsigned long old_len, unsigned long new_len,
1191                         unsigned long flags, unsigned long new_addr)
1192 {
1193         struct vm_area_struct *vma;
1194
1195         /* insanity checks first */
1196         if (new_len == 0)
1197                 return (unsigned long) -EINVAL;
1198
1199         if (flags & MREMAP_FIXED && new_addr != addr)
1200                 return (unsigned long) -EINVAL;
1201
1202         vma = find_vma_exact(current->mm, addr);
1203         if (!vma)
1204                 return (unsigned long) -EINVAL;
1205
1206         if (vma->vm_end != vma->vm_start + old_len)
1207                 return (unsigned long) -EFAULT;
1208
1209         if (vma->vm_flags & VM_MAYSHARE)
1210                 return (unsigned long) -EPERM;
1211
1212         if (new_len > kobjsize((void *) addr))
1213                 return (unsigned long) -ENOMEM;
1214
1215         /* all checks complete - do it */
1216         vma->vm_end = vma->vm_start + new_len;
1217
1218         askedalloc -= old_len;
1219         askedalloc += new_len;
1220
1221         return vma->vm_start;
1222 }
1223 EXPORT_SYMBOL(do_mremap);
1224
1225 asmlinkage unsigned long sys_mremap(unsigned long addr,
1226         unsigned long old_len, unsigned long new_len,
1227         unsigned long flags, unsigned long new_addr)
1228 {
1229         unsigned long ret;
1230
1231         down_write(&current->mm->mmap_sem);
1232         ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1233         up_write(&current->mm->mmap_sem);
1234         return ret;
1235 }
1236
1237 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1238                         unsigned int foll_flags)
1239 {
1240         return NULL;
1241 }
1242
1243 int remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
1244                 unsigned long to, unsigned long size, pgprot_t prot)
1245 {
1246         vma->vm_start = vma->vm_pgoff << PAGE_SHIFT;
1247         return 0;
1248 }
1249 EXPORT_SYMBOL(remap_pfn_range);
1250
1251 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1252                         unsigned long pgoff)
1253 {
1254         unsigned int size = vma->vm_end - vma->vm_start;
1255
1256         if (!(vma->vm_flags & VM_USERMAP))
1257                 return -EINVAL;
1258
1259         vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1260         vma->vm_end = vma->vm_start + size;
1261
1262         return 0;
1263 }
1264 EXPORT_SYMBOL(remap_vmalloc_range);
1265
1266 void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1267 {
1268 }
1269
1270 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1271         unsigned long len, unsigned long pgoff, unsigned long flags)
1272 {
1273         return -ENOMEM;
1274 }
1275
1276 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1277 {
1278 }
1279
1280 void unmap_mapping_range(struct address_space *mapping,
1281                          loff_t const holebegin, loff_t const holelen,
1282                          int even_cows)
1283 {
1284 }
1285 EXPORT_SYMBOL(unmap_mapping_range);
1286
1287 /*
1288  * ask for an unmapped area at which to create a mapping on a file
1289  */
1290 unsigned long get_unmapped_area(struct file *file, unsigned long addr,
1291                                 unsigned long len, unsigned long pgoff,
1292                                 unsigned long flags)
1293 {
1294         unsigned long (*get_area)(struct file *, unsigned long, unsigned long,
1295                                   unsigned long, unsigned long);
1296
1297         get_area = current->mm->get_unmapped_area;
1298         if (file && file->f_op && file->f_op->get_unmapped_area)
1299                 get_area = file->f_op->get_unmapped_area;
1300
1301         if (!get_area)
1302                 return -ENOSYS;
1303
1304         return get_area(file, addr, len, pgoff, flags);
1305 }
1306 EXPORT_SYMBOL(get_unmapped_area);
1307
1308 /*
1309  * Check that a process has enough memory to allocate a new virtual
1310  * mapping. 0 means there is enough memory for the allocation to
1311  * succeed and -ENOMEM implies there is not.
1312  *
1313  * We currently support three overcommit policies, which are set via the
1314  * vm.overcommit_memory sysctl.  See Documentation/vm/overcommit-accounting
1315  *
1316  * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1317  * Additional code 2002 Jul 20 by Robert Love.
1318  *
1319  * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1320  *
1321  * Note this is a helper function intended to be used by LSMs which
1322  * wish to use this logic.
1323  */
1324 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
1325 {
1326         unsigned long free, allowed;
1327
1328         vm_acct_memory(pages);
1329
1330         /*
1331          * Sometimes we want to use more memory than we have
1332          */
1333         if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1334                 return 0;
1335
1336         if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1337                 unsigned long n;
1338
1339                 free = global_page_state(NR_FILE_PAGES);
1340                 free += nr_swap_pages;
1341
1342                 /*
1343                  * Any slabs which are created with the
1344                  * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1345                  * which are reclaimable, under pressure.  The dentry
1346                  * cache and most inode caches should fall into this
1347                  */
1348                 free += global_page_state(NR_SLAB_RECLAIMABLE);
1349
1350                 /*
1351                  * Leave the last 3% for root
1352                  */
1353                 if (!cap_sys_admin)
1354                         free -= free / 32;
1355
1356                 if (free > pages)
1357                         return 0;
1358
1359                 /*
1360                  * nr_free_pages() is very expensive on large systems,
1361                  * only call if we're about to fail.
1362                  */
1363                 n = nr_free_pages();
1364
1365                 /*
1366                  * Leave reserved pages. The pages are not for anonymous pages.
1367                  */
1368                 if (n <= totalreserve_pages)
1369                         goto error;
1370                 else
1371                         n -= totalreserve_pages;
1372
1373                 /*
1374                  * Leave the last 3% for root
1375                  */
1376                 if (!cap_sys_admin)
1377                         n -= n / 32;
1378                 free += n;
1379
1380                 if (free > pages)
1381                         return 0;
1382
1383                 goto error;
1384         }
1385
1386         allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1387         /*
1388          * Leave the last 3% for root
1389          */
1390         if (!cap_sys_admin)
1391                 allowed -= allowed / 32;
1392         allowed += total_swap_pages;
1393
1394         /* Don't let a single process grow too big:
1395            leave 3% of the size of this process for other processes */
1396         allowed -= current->mm->total_vm / 32;
1397
1398         /*
1399          * cast `allowed' as a signed long because vm_committed_space
1400          * sometimes has a negative value
1401          */
1402         if (atomic_read(&vm_committed_space) < (long)allowed)
1403                 return 0;
1404 error:
1405         vm_unacct_memory(pages);
1406
1407         return -ENOMEM;
1408 }
1409
1410 int in_gate_area_no_task(unsigned long addr)
1411 {
1412         return 0;
1413 }
1414
1415 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1416 {
1417         BUG();
1418         return 0;
1419 }
1420 EXPORT_SYMBOL(filemap_fault);
1421
1422 /*
1423  * Access another process' address space.
1424  * - source/target buffer must be kernel space
1425  */
1426 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
1427 {
1428         struct vm_area_struct *vma;
1429         struct mm_struct *mm;
1430
1431         if (addr + len < addr)
1432                 return 0;
1433
1434         mm = get_task_mm(tsk);
1435         if (!mm)
1436                 return 0;
1437
1438         down_read(&mm->mmap_sem);
1439
1440         /* the access must start within one of the target process's mappings */
1441         vma = find_vma(mm, addr);
1442         if (vma) {
1443                 /* don't overrun this mapping */
1444                 if (addr + len >= vma->vm_end)
1445                         len = vma->vm_end - addr;
1446
1447                 /* only read or write mappings where it is permitted */
1448                 if (write && vma->vm_flags & VM_MAYWRITE)
1449                         len -= copy_to_user((void *) addr, buf, len);
1450                 else if (!write && vma->vm_flags & VM_MAYREAD)
1451                         len -= copy_from_user(buf, (void *) addr, len);
1452                 else
1453                         len = 0;
1454         } else {
1455                 len = 0;
1456         }
1457
1458         up_read(&mm->mmap_sem);
1459         mmput(mm);
1460         return len;
1461 }