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