4f87b2f43a2bd186a850e5f4136923f8e9b65eb4
[linux-2.6.git] / mm / nommu.c
1 /*
2  *  linux/mm/nommu.c
3  *
4  *  Replacement code for mm functions to support CPU's that don't
5  *  have any form of memory management unit (thus no virtual memory).
6  *
7  *  See Documentation/nommu-mmap.txt
8  *
9  *  Copyright (c) 2004-2005 David Howells <dhowells@redhat.com>
10  *  Copyright (c) 2000-2003 David McCullough <davidm@snapgear.com>
11  *  Copyright (c) 2000-2001 D Jeff Dionne <jeff@uClinux.org>
12  *  Copyright (c) 2002      Greg Ungerer <gerg@snapgear.com>
13  */
14
15 #include <linux/mm.h>
16 #include <linux/mman.h>
17 #include <linux/swap.h>
18 #include <linux/file.h>
19 #include <linux/highmem.h>
20 #include <linux/pagemap.h>
21 #include <linux/slab.h>
22 #include <linux/vmalloc.h>
23 #include <linux/ptrace.h>
24 #include <linux/blkdev.h>
25 #include <linux/backing-dev.h>
26 #include <linux/mount.h>
27 #include <linux/personality.h>
28 #include <linux/security.h>
29 #include <linux/syscalls.h>
30
31 #include <asm/uaccess.h>
32 #include <asm/tlb.h>
33 #include <asm/tlbflush.h>
34
35 void *high_memory;
36 struct page *mem_map;
37 unsigned long max_mapnr;
38 unsigned long num_physpages;
39 unsigned long askedalloc, realalloc;
40 atomic_t vm_committed_space = ATOMIC_INIT(0);
41 int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
42 int sysctl_overcommit_ratio = 50; /* default is 50% */
43 int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
44 int heap_stack_gap = 0;
45
46 EXPORT_SYMBOL(mem_map);
47 EXPORT_SYMBOL(__vm_enough_memory);
48
49 /* list of shareable VMAs */
50 struct rb_root nommu_vma_tree = RB_ROOT;
51 DECLARE_RWSEM(nommu_vma_sem);
52
53 struct vm_operations_struct generic_file_vm_ops = {
54 };
55
56 EXPORT_SYMBOL(vfree);
57 EXPORT_SYMBOL(vmalloc_to_page);
58 EXPORT_SYMBOL(vmalloc_32);
59 EXPORT_SYMBOL(vmap);
60 EXPORT_SYMBOL(vunmap);
61
62 /*
63  * Handle all mappings that got truncated by a "truncate()"
64  * system call.
65  *
66  * NOTE! We have to be ready to update the memory sharing
67  * between the file and the memory map for a potential last
68  * incomplete page.  Ugly, but necessary.
69  */
70 int vmtruncate(struct inode *inode, loff_t offset)
71 {
72         struct address_space *mapping = inode->i_mapping;
73         unsigned long limit;
74
75         if (inode->i_size < offset)
76                 goto do_expand;
77         i_size_write(inode, offset);
78
79         truncate_inode_pages(mapping, offset);
80         goto out_truncate;
81
82 do_expand:
83         limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
84         if (limit != RLIM_INFINITY && offset > limit)
85                 goto out_sig;
86         if (offset > inode->i_sb->s_maxbytes)
87                 goto out;
88         i_size_write(inode, offset);
89
90 out_truncate:
91         if (inode->i_op && inode->i_op->truncate)
92                 inode->i_op->truncate(inode);
93         return 0;
94 out_sig:
95         send_sig(SIGXFSZ, current, 0);
96 out:
97         return -EFBIG;
98 }
99
100 EXPORT_SYMBOL(vmtruncate);
101
102 /*
103  * Return the total memory allocated for this pointer, not
104  * just what the caller asked for.
105  *
106  * Doesn't have to be accurate, i.e. may have races.
107  */
108 unsigned int kobjsize(const void *objp)
109 {
110         struct page *page;
111
112         if (!objp || !((page = virt_to_page(objp))))
113                 return 0;
114
115         if (PageSlab(page))
116                 return ksize(objp);
117
118         BUG_ON(page->index < 0);
119         BUG_ON(page->index >= MAX_ORDER);
120
121         return (PAGE_SIZE << page->index);
122 }
123
124 /*
125  * get a list of pages in an address range belonging to the specified process
126  * and indicate the VMA that covers each page
127  * - this is potentially dodgy as we may end incrementing the page count of a
128  *   slab page or a secondary page from a compound page
129  * - don't permit access to VMAs that don't support it, such as I/O mappings
130  */
131 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
132         unsigned long start, int len, int write, int force,
133         struct page **pages, struct vm_area_struct **vmas)
134 {
135         struct vm_area_struct *vma;
136         unsigned long vm_flags;
137         int i;
138
139         /* calculate required read or write permissions.
140          * - if 'force' is set, we only require the "MAY" flags.
141          */
142         vm_flags  = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
143         vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
144
145         for (i = 0; i < len; i++) {
146                 vma = find_vma(mm, start);
147                 if (!vma)
148                         goto finish_or_fault;
149
150                 /* protect what we can, including chardevs */
151                 if (vma->vm_flags & (VM_IO | VM_PFNMAP) ||
152                     !(vm_flags & vma->vm_flags))
153                         goto finish_or_fault;
154
155                 if (pages) {
156                         pages[i] = virt_to_page(start);
157                         if (pages[i])
158                                 page_cache_get(pages[i]);
159                 }
160                 if (vmas)
161                         vmas[i] = vma;
162                 start += PAGE_SIZE;
163         }
164
165         return i;
166
167 finish_or_fault:
168         return i ? : -EFAULT;
169 }
170
171 EXPORT_SYMBOL(get_user_pages);
172
173 DEFINE_RWLOCK(vmlist_lock);
174 struct vm_struct *vmlist;
175
176 void vfree(void *addr)
177 {
178         kfree(addr);
179 }
180
181 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
182 {
183         /*
184          * kmalloc doesn't like __GFP_HIGHMEM for some reason
185          */
186         return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
187 }
188
189 struct page * vmalloc_to_page(void *addr)
190 {
191         return virt_to_page(addr);
192 }
193
194 unsigned long vmalloc_to_pfn(void *addr)
195 {
196         return page_to_pfn(virt_to_page(addr));
197 }
198
199
200 long vread(char *buf, char *addr, unsigned long count)
201 {
202         memcpy(buf, addr, count);
203         return count;
204 }
205
206 long vwrite(char *buf, char *addr, unsigned long count)
207 {
208         /* Don't allow overflow */
209         if ((unsigned long) addr + count < count)
210                 count = -(unsigned long) addr;
211
212         memcpy(addr, buf, count);
213         return(count);
214 }
215
216 /*
217  *      vmalloc  -  allocate virtually continguos memory
218  *
219  *      @size:          allocation size
220  *
221  *      Allocate enough pages to cover @size from the page level
222  *      allocator and map them into continguos kernel virtual space.
223  *
224  *      For tight cotrol over page level allocator and protection flags
225  *      use __vmalloc() instead.
226  */
227 void *vmalloc(unsigned long size)
228 {
229        return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
230 }
231 EXPORT_SYMBOL(vmalloc);
232
233 void *vmalloc_node(unsigned long size, int node)
234 {
235         return vmalloc(size);
236 }
237 EXPORT_SYMBOL(vmalloc_node);
238
239 /*
240  *      vmalloc_32  -  allocate virtually continguos memory (32bit addressable)
241  *
242  *      @size:          allocation size
243  *
244  *      Allocate enough 32bit PA addressable pages to cover @size from the
245  *      page level allocator and map them into continguos kernel virtual space.
246  */
247 void *vmalloc_32(unsigned long size)
248 {
249         return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
250 }
251
252 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
253 {
254         BUG();
255         return NULL;
256 }
257
258 void vunmap(void *addr)
259 {
260         BUG();
261 }
262
263 /*
264  *  sys_brk() for the most part doesn't need the global kernel
265  *  lock, except when an application is doing something nasty
266  *  like trying to un-brk an area that has already been mapped
267  *  to a regular file.  in this case, the unmapping will need
268  *  to invoke file system routines that need the global lock.
269  */
270 asmlinkage unsigned long sys_brk(unsigned long brk)
271 {
272         struct mm_struct *mm = current->mm;
273
274         if (brk < mm->start_brk || brk > mm->context.end_brk)
275                 return mm->brk;
276
277         if (mm->brk == brk)
278                 return mm->brk;
279
280         /*
281          * Always allow shrinking brk
282          */
283         if (brk <= mm->brk) {
284                 mm->brk = brk;
285                 return brk;
286         }
287
288         /*
289          * Ok, looks good - let it rip.
290          */
291         return mm->brk = brk;
292 }
293
294 #ifdef DEBUG
295 static void show_process_blocks(void)
296 {
297         struct vm_list_struct *vml;
298
299         printk("Process blocks %d:", current->pid);
300
301         for (vml = &current->mm->context.vmlist; vml; vml = vml->next) {
302                 printk(" %p: %p", vml, vml->vma);
303                 if (vml->vma)
304                         printk(" (%d @%lx #%d)",
305                                kobjsize((void *) vml->vma->vm_start),
306                                vml->vma->vm_start,
307                                atomic_read(&vml->vma->vm_usage));
308                 printk(vml->next ? " ->" : ".\n");
309         }
310 }
311 #endif /* DEBUG */
312
313 /*
314  * add a VMA into a process's mm_struct in the appropriate place in the list
315  * - should be called with mm->mmap_sem held writelocked
316  */
317 static void add_vma_to_mm(struct mm_struct *mm, struct vm_list_struct *vml)
318 {
319         struct vm_list_struct **ppv;
320
321         for (ppv = &current->mm->context.vmlist; *ppv; ppv = &(*ppv)->next)
322                 if ((*ppv)->vma->vm_start > vml->vma->vm_start)
323                         break;
324
325         vml->next = *ppv;
326         *ppv = vml;
327 }
328
329 /*
330  * look up the first VMA in which addr resides, NULL if none
331  * - should be called with mm->mmap_sem at least held readlocked
332  */
333 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
334 {
335         struct vm_list_struct *loop, *vml;
336
337         /* search the vm_start ordered list */
338         vml = NULL;
339         for (loop = mm->context.vmlist; loop; loop = loop->next) {
340                 if (loop->vma->vm_start > addr)
341                         break;
342                 vml = loop;
343         }
344
345         if (vml && vml->vma->vm_end > addr)
346                 return vml->vma;
347
348         return NULL;
349 }
350 EXPORT_SYMBOL(find_vma);
351
352 /*
353  * find a VMA in the global tree
354  */
355 static inline struct vm_area_struct *find_nommu_vma(unsigned long start)
356 {
357         struct vm_area_struct *vma;
358         struct rb_node *n = nommu_vma_tree.rb_node;
359
360         while (n) {
361                 vma = rb_entry(n, struct vm_area_struct, vm_rb);
362
363                 if (start < vma->vm_start)
364                         n = n->rb_left;
365                 else if (start > vma->vm_start)
366                         n = n->rb_right;
367                 else
368                         return vma;
369         }
370
371         return NULL;
372 }
373
374 /*
375  * add a VMA in the global tree
376  */
377 static void add_nommu_vma(struct vm_area_struct *vma)
378 {
379         struct vm_area_struct *pvma;
380         struct address_space *mapping;
381         struct rb_node **p = &nommu_vma_tree.rb_node;
382         struct rb_node *parent = NULL;
383
384         /* add the VMA to the mapping */
385         if (vma->vm_file) {
386                 mapping = vma->vm_file->f_mapping;
387
388                 flush_dcache_mmap_lock(mapping);
389                 vma_prio_tree_insert(vma, &mapping->i_mmap);
390                 flush_dcache_mmap_unlock(mapping);
391         }
392
393         /* add the VMA to the master list */
394         while (*p) {
395                 parent = *p;
396                 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
397
398                 if (vma->vm_start < pvma->vm_start) {
399                         p = &(*p)->rb_left;
400                 }
401                 else if (vma->vm_start > pvma->vm_start) {
402                         p = &(*p)->rb_right;
403                 }
404                 else {
405                         /* mappings are at the same address - this can only
406                          * happen for shared-mem chardevs and shared file
407                          * mappings backed by ramfs/tmpfs */
408                         BUG_ON(!(pvma->vm_flags & VM_SHARED));
409
410                         if (vma < pvma)
411                                 p = &(*p)->rb_left;
412                         else if (vma > pvma)
413                                 p = &(*p)->rb_right;
414                         else
415                                 BUG();
416                 }
417         }
418
419         rb_link_node(&vma->vm_rb, parent, p);
420         rb_insert_color(&vma->vm_rb, &nommu_vma_tree);
421 }
422
423 /*
424  * delete a VMA from the global list
425  */
426 static void delete_nommu_vma(struct vm_area_struct *vma)
427 {
428         struct address_space *mapping;
429
430         /* remove the VMA from the mapping */
431         if (vma->vm_file) {
432                 mapping = vma->vm_file->f_mapping;
433
434                 flush_dcache_mmap_lock(mapping);
435                 vma_prio_tree_remove(vma, &mapping->i_mmap);
436                 flush_dcache_mmap_unlock(mapping);
437         }
438
439         /* remove from the master list */
440         rb_erase(&vma->vm_rb, &nommu_vma_tree);
441 }
442
443 /*
444  * determine whether a mapping should be permitted and, if so, what sort of
445  * mapping we're capable of supporting
446  */
447 static int validate_mmap_request(struct file *file,
448                                  unsigned long addr,
449                                  unsigned long len,
450                                  unsigned long prot,
451                                  unsigned long flags,
452                                  unsigned long pgoff,
453                                  unsigned long *_capabilities)
454 {
455         unsigned long capabilities;
456         unsigned long reqprot = prot;
457         int ret;
458
459         /* do the simple checks first */
460         if (flags & MAP_FIXED || addr) {
461                 printk(KERN_DEBUG
462                        "%d: Can't do fixed-address/overlay mmap of RAM\n",
463                        current->pid);
464                 return -EINVAL;
465         }
466
467         if ((flags & MAP_TYPE) != MAP_PRIVATE &&
468             (flags & MAP_TYPE) != MAP_SHARED)
469                 return -EINVAL;
470
471         if (PAGE_ALIGN(len) == 0)
472                 return addr;
473
474         if (len > TASK_SIZE)
475                 return -EINVAL;
476
477         /* offset overflow? */
478         if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
479                 return -EINVAL;
480
481         if (file) {
482                 /* validate file mapping requests */
483                 struct address_space *mapping;
484
485                 /* files must support mmap */
486                 if (!file->f_op || !file->f_op->mmap)
487                         return -ENODEV;
488
489                 /* work out if what we've got could possibly be shared
490                  * - we support chardevs that provide their own "memory"
491                  * - we support files/blockdevs that are memory backed
492                  */
493                 mapping = file->f_mapping;
494                 if (!mapping)
495                         mapping = file->f_dentry->d_inode->i_mapping;
496
497                 capabilities = 0;
498                 if (mapping && mapping->backing_dev_info)
499                         capabilities = mapping->backing_dev_info->capabilities;
500
501                 if (!capabilities) {
502                         /* no explicit capabilities set, so assume some
503                          * defaults */
504                         switch (file->f_dentry->d_inode->i_mode & S_IFMT) {
505                         case S_IFREG:
506                         case S_IFBLK:
507                                 capabilities = BDI_CAP_MAP_COPY;
508                                 break;
509
510                         case S_IFCHR:
511                                 capabilities =
512                                         BDI_CAP_MAP_DIRECT |
513                                         BDI_CAP_READ_MAP |
514                                         BDI_CAP_WRITE_MAP;
515                                 break;
516
517                         default:
518                                 return -EINVAL;
519                         }
520                 }
521
522                 /* eliminate any capabilities that we can't support on this
523                  * device */
524                 if (!file->f_op->get_unmapped_area)
525                         capabilities &= ~BDI_CAP_MAP_DIRECT;
526                 if (!file->f_op->read)
527                         capabilities &= ~BDI_CAP_MAP_COPY;
528
529                 if (flags & MAP_SHARED) {
530                         /* do checks for writing, appending and locking */
531                         if ((prot & PROT_WRITE) &&
532                             !(file->f_mode & FMODE_WRITE))
533                                 return -EACCES;
534
535                         if (IS_APPEND(file->f_dentry->d_inode) &&
536                             (file->f_mode & FMODE_WRITE))
537                                 return -EACCES;
538
539                         if (locks_verify_locked(file->f_dentry->d_inode))
540                                 return -EAGAIN;
541
542                         if (!(capabilities & BDI_CAP_MAP_DIRECT))
543                                 return -ENODEV;
544
545                         if (((prot & PROT_READ)  && !(capabilities & BDI_CAP_READ_MAP))  ||
546                             ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
547                             ((prot & PROT_EXEC)  && !(capabilities & BDI_CAP_EXEC_MAP))
548                             ) {
549                                 printk("MAP_SHARED not completely supported on !MMU\n");
550                                 return -EINVAL;
551                         }
552
553                         /* we mustn't privatise shared mappings */
554                         capabilities &= ~BDI_CAP_MAP_COPY;
555                 }
556                 else {
557                         /* we're going to read the file into private memory we
558                          * allocate */
559                         if (!(capabilities & BDI_CAP_MAP_COPY))
560                                 return -ENODEV;
561
562                         /* we don't permit a private writable mapping to be
563                          * shared with the backing device */
564                         if (prot & PROT_WRITE)
565                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
566                 }
567
568                 /* handle executable mappings and implied executable
569                  * mappings */
570                 if (file->f_vfsmnt->mnt_flags & MNT_NOEXEC) {
571                         if (prot & PROT_EXEC)
572                                 return -EPERM;
573                 }
574                 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
575                         /* handle implication of PROT_EXEC by PROT_READ */
576                         if (current->personality & READ_IMPLIES_EXEC) {
577                                 if (capabilities & BDI_CAP_EXEC_MAP)
578                                         prot |= PROT_EXEC;
579                         }
580                 }
581                 else if ((prot & PROT_READ) &&
582                          (prot & PROT_EXEC) &&
583                          !(capabilities & BDI_CAP_EXEC_MAP)
584                          ) {
585                         /* backing file is not executable, try to copy */
586                         capabilities &= ~BDI_CAP_MAP_DIRECT;
587                 }
588         }
589         else {
590                 /* anonymous mappings are always memory backed and can be
591                  * privately mapped
592                  */
593                 capabilities = BDI_CAP_MAP_COPY;
594
595                 /* handle PROT_EXEC implication by PROT_READ */
596                 if ((prot & PROT_READ) &&
597                     (current->personality & READ_IMPLIES_EXEC))
598                         prot |= PROT_EXEC;
599         }
600
601         /* allow the security API to have its say */
602         ret = security_file_mmap(file, reqprot, prot, flags);
603         if (ret < 0)
604                 return ret;
605
606         /* looks okay */
607         *_capabilities = capabilities;
608         return 0;
609 }
610
611 /*
612  * we've determined that we can make the mapping, now translate what we
613  * now know into VMA flags
614  */
615 static unsigned long determine_vm_flags(struct file *file,
616                                         unsigned long prot,
617                                         unsigned long flags,
618                                         unsigned long capabilities)
619 {
620         unsigned long vm_flags;
621
622         vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
623         vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
624         /* vm_flags |= mm->def_flags; */
625
626         if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
627                 /* attempt to share read-only copies of mapped file chunks */
628                 if (file && !(prot & PROT_WRITE))
629                         vm_flags |= VM_MAYSHARE;
630         }
631         else {
632                 /* overlay a shareable mapping on the backing device or inode
633                  * if possible - used for chardevs, ramfs/tmpfs/shmfs and
634                  * romfs/cramfs */
635                 if (flags & MAP_SHARED)
636                         vm_flags |= VM_MAYSHARE | VM_SHARED;
637                 else if ((((vm_flags & capabilities) ^ vm_flags) & BDI_CAP_VMFLAGS) == 0)
638                         vm_flags |= VM_MAYSHARE;
639         }
640
641         /* refuse to let anyone share private mappings with this process if
642          * it's being traced - otherwise breakpoints set in it may interfere
643          * with another untraced process
644          */
645         if ((flags & MAP_PRIVATE) && (current->ptrace & PT_PTRACED))
646                 vm_flags &= ~VM_MAYSHARE;
647
648         return vm_flags;
649 }
650
651 /*
652  * set up a shared mapping on a file
653  */
654 static int do_mmap_shared_file(struct vm_area_struct *vma, unsigned long len)
655 {
656         int ret;
657
658         ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
659         if (ret != -ENOSYS)
660                 return ret;
661
662         /* getting an ENOSYS error indicates that direct mmap isn't
663          * possible (as opposed to tried but failed) so we'll fall
664          * through to making a private copy of the data and mapping
665          * that if we can */
666         return -ENODEV;
667 }
668
669 /*
670  * set up a private mapping or an anonymous shared mapping
671  */
672 static int do_mmap_private(struct vm_area_struct *vma, unsigned long len)
673 {
674         void *base;
675         int ret;
676
677         /* invoke the file's mapping function so that it can keep track of
678          * shared mappings on devices or memory
679          * - VM_MAYSHARE will be set if it may attempt to share
680          */
681         if (vma->vm_file) {
682                 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
683                 if (ret != -ENOSYS) {
684                         /* shouldn't return success if we're not sharing */
685                         BUG_ON(ret == 0 && !(vma->vm_flags & VM_MAYSHARE));
686                         return ret; /* success or a real error */
687                 }
688
689                 /* getting an ENOSYS error indicates that direct mmap isn't
690                  * possible (as opposed to tried but failed) so we'll try to
691                  * make a private copy of the data and map that instead */
692         }
693
694         /* allocate some memory to hold the mapping
695          * - note that this may not return a page-aligned address if the object
696          *   we're allocating is smaller than a page
697          */
698         base = kmalloc(len, GFP_KERNEL|__GFP_COMP);
699         if (!base)
700                 goto enomem;
701
702         vma->vm_start = (unsigned long) base;
703         vma->vm_end = vma->vm_start + len;
704         vma->vm_flags |= VM_MAPPED_COPY;
705
706 #ifdef WARN_ON_SLACK
707         if (len + WARN_ON_SLACK <= kobjsize(result))
708                 printk("Allocation of %lu bytes from process %d has %lu bytes of slack\n",
709                        len, current->pid, kobjsize(result) - len);
710 #endif
711
712         if (vma->vm_file) {
713                 /* read the contents of a file into the copy */
714                 mm_segment_t old_fs;
715                 loff_t fpos;
716
717                 fpos = vma->vm_pgoff;
718                 fpos <<= PAGE_SHIFT;
719
720                 old_fs = get_fs();
721                 set_fs(KERNEL_DS);
722                 ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos);
723                 set_fs(old_fs);
724
725                 if (ret < 0)
726                         goto error_free;
727
728                 /* clear the last little bit */
729                 if (ret < len)
730                         memset(base + ret, 0, len - ret);
731
732         } else {
733                 /* if it's an anonymous mapping, then just clear it */
734                 memset(base, 0, len);
735         }
736
737         return 0;
738
739 error_free:
740         kfree(base);
741         vma->vm_start = 0;
742         return ret;
743
744 enomem:
745         printk("Allocation of length %lu from process %d failed\n",
746                len, current->pid);
747         show_free_areas();
748         return -ENOMEM;
749 }
750
751 /*
752  * handle mapping creation for uClinux
753  */
754 unsigned long do_mmap_pgoff(struct file *file,
755                             unsigned long addr,
756                             unsigned long len,
757                             unsigned long prot,
758                             unsigned long flags,
759                             unsigned long pgoff)
760 {
761         struct vm_list_struct *vml = NULL;
762         struct vm_area_struct *vma = NULL;
763         struct rb_node *rb;
764         unsigned long capabilities, vm_flags;
765         void *result;
766         int ret;
767
768         /* decide whether we should attempt the mapping, and if so what sort of
769          * mapping */
770         ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
771                                     &capabilities);
772         if (ret < 0)
773                 return ret;
774
775         /* we've determined that we can make the mapping, now translate what we
776          * now know into VMA flags */
777         vm_flags = determine_vm_flags(file, prot, flags, capabilities);
778
779         /* we're going to need to record the mapping if it works */
780         vml = kmalloc(sizeof(struct vm_list_struct), GFP_KERNEL);
781         if (!vml)
782                 goto error_getting_vml;
783         memset(vml, 0, sizeof(*vml));
784
785         down_write(&nommu_vma_sem);
786
787         /* if we want to share, we need to check for VMAs created by other
788          * mmap() calls that overlap with our proposed mapping
789          * - we can only share with an exact match on most regular files
790          * - shared mappings on character devices and memory backed files are
791          *   permitted to overlap inexactly as far as we are concerned for in
792          *   these cases, sharing is handled in the driver or filesystem rather
793          *   than here
794          */
795         if (vm_flags & VM_MAYSHARE) {
796                 unsigned long pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
797                 unsigned long vmpglen;
798
799                 for (rb = rb_first(&nommu_vma_tree); rb; rb = rb_next(rb)) {
800                         vma = rb_entry(rb, struct vm_area_struct, vm_rb);
801
802                         if (!(vma->vm_flags & VM_MAYSHARE))
803                                 continue;
804
805                         /* search for overlapping mappings on the same file */
806                         if (vma->vm_file->f_dentry->d_inode != file->f_dentry->d_inode)
807                                 continue;
808
809                         if (vma->vm_pgoff >= pgoff + pglen)
810                                 continue;
811
812                         vmpglen = vma->vm_end - vma->vm_start + PAGE_SIZE - 1;
813                         vmpglen >>= PAGE_SHIFT;
814                         if (pgoff >= vma->vm_pgoff + vmpglen)
815                                 continue;
816
817                         /* handle inexactly overlapping matches between mappings */
818                         if (vma->vm_pgoff != pgoff || vmpglen != pglen) {
819                                 if (!(capabilities & BDI_CAP_MAP_DIRECT))
820                                         goto sharing_violation;
821                                 continue;
822                         }
823
824                         /* we've found a VMA we can share */
825                         atomic_inc(&vma->vm_usage);
826
827                         vml->vma = vma;
828                         result = (void *) vma->vm_start;
829                         goto shared;
830                 }
831
832                 vma = NULL;
833
834                 /* obtain the address at which to make a shared mapping
835                  * - this is the hook for quasi-memory character devices to
836                  *   tell us the location of a shared mapping
837                  */
838                 if (file && file->f_op->get_unmapped_area) {
839                         addr = file->f_op->get_unmapped_area(file, addr, len,
840                                                              pgoff, flags);
841                         if (IS_ERR((void *) addr)) {
842                                 ret = addr;
843                                 if (ret != (unsigned long) -ENOSYS)
844                                         goto error;
845
846                                 /* the driver refused to tell us where to site
847                                  * the mapping so we'll have to attempt to copy
848                                  * it */
849                                 ret = (unsigned long) -ENODEV;
850                                 if (!(capabilities & BDI_CAP_MAP_COPY))
851                                         goto error;
852
853                                 capabilities &= ~BDI_CAP_MAP_DIRECT;
854                         }
855                 }
856         }
857
858         /* we're going to need a VMA struct as well */
859         vma = kmalloc(sizeof(struct vm_area_struct), GFP_KERNEL);
860         if (!vma)
861                 goto error_getting_vma;
862
863         memset(vma, 0, sizeof(*vma));
864         INIT_LIST_HEAD(&vma->anon_vma_node);
865         atomic_set(&vma->vm_usage, 1);
866         if (file)
867                 get_file(file);
868         vma->vm_file    = file;
869         vma->vm_flags   = vm_flags;
870         vma->vm_start   = addr;
871         vma->vm_end     = addr + len;
872         vma->vm_pgoff   = pgoff;
873
874         vml->vma = vma;
875
876         /* set up the mapping */
877         if (file && vma->vm_flags & VM_SHARED)
878                 ret = do_mmap_shared_file(vma, len);
879         else
880                 ret = do_mmap_private(vma, len);
881         if (ret < 0)
882                 goto error;
883
884         /* okay... we have a mapping; now we have to register it */
885         result = (void *) vma->vm_start;
886
887         if (vma->vm_flags & VM_MAPPED_COPY) {
888                 realalloc += kobjsize(result);
889                 askedalloc += len;
890         }
891
892         realalloc += kobjsize(vma);
893         askedalloc += sizeof(*vma);
894
895         current->mm->total_vm += len >> PAGE_SHIFT;
896
897         add_nommu_vma(vma);
898
899  shared:
900         realalloc += kobjsize(vml);
901         askedalloc += sizeof(*vml);
902
903         add_vma_to_mm(current->mm, vml);
904
905         up_write(&nommu_vma_sem);
906
907         if (prot & PROT_EXEC)
908                 flush_icache_range((unsigned long) result,
909                                    (unsigned long) result + len);
910
911 #ifdef DEBUG
912         printk("do_mmap:\n");
913         show_process_blocks();
914 #endif
915
916         return (unsigned long) result;
917
918  error:
919         up_write(&nommu_vma_sem);
920         kfree(vml);
921         if (vma) {
922                 fput(vma->vm_file);
923                 kfree(vma);
924         }
925         return ret;
926
927  sharing_violation:
928         up_write(&nommu_vma_sem);
929         printk("Attempt to share mismatched mappings\n");
930         kfree(vml);
931         return -EINVAL;
932
933  error_getting_vma:
934         up_write(&nommu_vma_sem);
935         kfree(vml);
936         printk("Allocation of vma for %lu byte allocation from process %d failed\n",
937                len, current->pid);
938         show_free_areas();
939         return -ENOMEM;
940
941  error_getting_vml:
942         printk("Allocation of vml for %lu byte allocation from process %d failed\n",
943                len, current->pid);
944         show_free_areas();
945         return -ENOMEM;
946 }
947
948 /*
949  * handle mapping disposal for uClinux
950  */
951 static void put_vma(struct vm_area_struct *vma)
952 {
953         if (vma) {
954                 down_write(&nommu_vma_sem);
955
956                 if (atomic_dec_and_test(&vma->vm_usage)) {
957                         delete_nommu_vma(vma);
958
959                         if (vma->vm_ops && vma->vm_ops->close)
960                                 vma->vm_ops->close(vma);
961
962                         /* IO memory and memory shared directly out of the pagecache from
963                          * ramfs/tmpfs mustn't be released here */
964                         if (vma->vm_flags & VM_MAPPED_COPY) {
965                                 realalloc -= kobjsize((void *) vma->vm_start);
966                                 askedalloc -= vma->vm_end - vma->vm_start;
967                                 kfree((void *) vma->vm_start);
968                         }
969
970                         realalloc -= kobjsize(vma);
971                         askedalloc -= sizeof(*vma);
972
973                         if (vma->vm_file)
974                                 fput(vma->vm_file);
975                         kfree(vma);
976                 }
977
978                 up_write(&nommu_vma_sem);
979         }
980 }
981
982 /*
983  * release a mapping
984  * - under NOMMU conditions the parameters must match exactly to the mapping to
985  *   be removed
986  */
987 int do_munmap(struct mm_struct *mm, unsigned long addr, size_t len)
988 {
989         struct vm_list_struct *vml, **parent;
990         unsigned long end = addr + len;
991
992 #ifdef DEBUG
993         printk("do_munmap:\n");
994 #endif
995
996         for (parent = &mm->context.vmlist; *parent; parent = &(*parent)->next) {
997                 if ((*parent)->vma->vm_start > addr)
998                         break;
999                 if ((*parent)->vma->vm_start == addr &&
1000                     ((len == 0) || ((*parent)->vma->vm_end == end)))
1001                         goto found;
1002         }
1003
1004         printk("munmap of non-mmaped memory by process %d (%s): %p\n",
1005                current->pid, current->comm, (void *) addr);
1006         return -EINVAL;
1007
1008  found:
1009         vml = *parent;
1010
1011         put_vma(vml->vma);
1012
1013         *parent = vml->next;
1014         realalloc -= kobjsize(vml);
1015         askedalloc -= sizeof(*vml);
1016         kfree(vml);
1017
1018         update_hiwater_vm(mm);
1019         mm->total_vm -= len >> PAGE_SHIFT;
1020
1021 #ifdef DEBUG
1022         show_process_blocks();
1023 #endif
1024
1025         return 0;
1026 }
1027
1028 asmlinkage long sys_munmap(unsigned long addr, size_t len)
1029 {
1030         int ret;
1031         struct mm_struct *mm = current->mm;
1032
1033         down_write(&mm->mmap_sem);
1034         ret = do_munmap(mm, addr, len);
1035         up_write(&mm->mmap_sem);
1036         return ret;
1037 }
1038
1039 /*
1040  * Release all mappings
1041  */
1042 void exit_mmap(struct mm_struct * mm)
1043 {
1044         struct vm_list_struct *tmp;
1045
1046         if (mm) {
1047 #ifdef DEBUG
1048                 printk("Exit_mmap:\n");
1049 #endif
1050
1051                 mm->total_vm = 0;
1052
1053                 while ((tmp = mm->context.vmlist)) {
1054                         mm->context.vmlist = tmp->next;
1055                         put_vma(tmp->vma);
1056
1057                         realalloc -= kobjsize(tmp);
1058                         askedalloc -= sizeof(*tmp);
1059                         kfree(tmp);
1060                 }
1061
1062 #ifdef DEBUG
1063                 show_process_blocks();
1064 #endif
1065         }
1066 }
1067
1068 unsigned long do_brk(unsigned long addr, unsigned long len)
1069 {
1070         return -ENOMEM;
1071 }
1072
1073 /*
1074  * Expand (or shrink) an existing mapping, potentially moving it at the
1075  * same time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1076  *
1077  * MREMAP_FIXED option added 5-Dec-1999 by Benjamin LaHaise
1078  * This option implies MREMAP_MAYMOVE.
1079  *
1080  * on uClinux, we only permit changing a mapping's size, and only as long as it stays within the
1081  * hole allocated by the kmalloc() call in do_mmap_pgoff() and the block is not shareable
1082  */
1083 unsigned long do_mremap(unsigned long addr,
1084                         unsigned long old_len, unsigned long new_len,
1085                         unsigned long flags, unsigned long new_addr)
1086 {
1087         struct vm_list_struct *vml = NULL;
1088
1089         /* insanity checks first */
1090         if (new_len == 0)
1091                 return (unsigned long) -EINVAL;
1092
1093         if (flags & MREMAP_FIXED && new_addr != addr)
1094                 return (unsigned long) -EINVAL;
1095
1096         for (vml = current->mm->context.vmlist; vml; vml = vml->next)
1097                 if (vml->vma->vm_start == addr)
1098                         goto found;
1099
1100         return (unsigned long) -EINVAL;
1101
1102  found:
1103         if (vml->vma->vm_end != vml->vma->vm_start + old_len)
1104                 return (unsigned long) -EFAULT;
1105
1106         if (vml->vma->vm_flags & VM_MAYSHARE)
1107                 return (unsigned long) -EPERM;
1108
1109         if (new_len > kobjsize((void *) addr))
1110                 return (unsigned long) -ENOMEM;
1111
1112         /* all checks complete - do it */
1113         vml->vma->vm_end = vml->vma->vm_start + new_len;
1114
1115         askedalloc -= old_len;
1116         askedalloc += new_len;
1117
1118         return vml->vma->vm_start;
1119 }
1120
1121 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1122                         unsigned int foll_flags)
1123 {
1124         return NULL;
1125 }
1126
1127 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
1128 {
1129         return NULL;
1130 }
1131
1132 int remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
1133                 unsigned long to, unsigned long size, pgprot_t prot)
1134 {
1135         vma->vm_start = vma->vm_pgoff << PAGE_SHIFT;
1136         return 0;
1137 }
1138 EXPORT_SYMBOL(remap_pfn_range);
1139
1140 void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1141 {
1142 }
1143
1144 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1145         unsigned long len, unsigned long pgoff, unsigned long flags)
1146 {
1147         return -ENOMEM;
1148 }
1149
1150 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1151 {
1152 }
1153
1154 void unmap_mapping_range(struct address_space *mapping,
1155                          loff_t const holebegin, loff_t const holelen,
1156                          int even_cows)
1157 {
1158 }
1159 EXPORT_SYMBOL(unmap_mapping_range);
1160
1161 /*
1162  * Check that a process has enough memory to allocate a new virtual
1163  * mapping. 0 means there is enough memory for the allocation to
1164  * succeed and -ENOMEM implies there is not.
1165  *
1166  * We currently support three overcommit policies, which are set via the
1167  * vm.overcommit_memory sysctl.  See Documentation/vm/overcommit-accounting
1168  *
1169  * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1170  * Additional code 2002 Jul 20 by Robert Love.
1171  *
1172  * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1173  *
1174  * Note this is a helper function intended to be used by LSMs which
1175  * wish to use this logic.
1176  */
1177 int __vm_enough_memory(long pages, int cap_sys_admin)
1178 {
1179         unsigned long free, allowed;
1180
1181         vm_acct_memory(pages);
1182
1183         /*
1184          * Sometimes we want to use more memory than we have
1185          */
1186         if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1187                 return 0;
1188
1189         if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1190                 unsigned long n;
1191
1192                 free = global_page_state(NR_FILE_PAGES);
1193                 free += nr_swap_pages;
1194
1195                 /*
1196                  * Any slabs which are created with the
1197                  * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1198                  * which are reclaimable, under pressure.  The dentry
1199                  * cache and most inode caches should fall into this
1200                  */
1201                 free += global_page_state(NR_SLAB_RECLAIMABLE);
1202
1203                 /*
1204                  * Leave the last 3% for root
1205                  */
1206                 if (!cap_sys_admin)
1207                         free -= free / 32;
1208
1209                 if (free > pages)
1210                         return 0;
1211
1212                 /*
1213                  * nr_free_pages() is very expensive on large systems,
1214                  * only call if we're about to fail.
1215                  */
1216                 n = nr_free_pages();
1217
1218                 /*
1219                  * Leave reserved pages. The pages are not for anonymous pages.
1220                  */
1221                 if (n <= totalreserve_pages)
1222                         goto error;
1223                 else
1224                         n -= totalreserve_pages;
1225
1226                 /*
1227                  * Leave the last 3% for root
1228                  */
1229                 if (!cap_sys_admin)
1230                         n -= n / 32;
1231                 free += n;
1232
1233                 if (free > pages)
1234                         return 0;
1235
1236                 goto error;
1237         }
1238
1239         allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1240         /*
1241          * Leave the last 3% for root
1242          */
1243         if (!cap_sys_admin)
1244                 allowed -= allowed / 32;
1245         allowed += total_swap_pages;
1246
1247         /* Don't let a single process grow too big:
1248            leave 3% of the size of this process for other processes */
1249         allowed -= current->mm->total_vm / 32;
1250
1251         /*
1252          * cast `allowed' as a signed long because vm_committed_space
1253          * sometimes has a negative value
1254          */
1255         if (atomic_read(&vm_committed_space) < (long)allowed)
1256                 return 0;
1257 error:
1258         vm_unacct_memory(pages);
1259
1260         return -ENOMEM;
1261 }
1262
1263 int in_gate_area_no_task(unsigned long addr)
1264 {
1265         return 0;
1266 }
1267
1268 struct page *filemap_nopage(struct vm_area_struct *area,
1269                         unsigned long address, int *type)
1270 {
1271         BUG();
1272         return NULL;
1273 }
1274
1275 /*
1276  * Access another process' address space.
1277  * - source/target buffer must be kernel space
1278  */
1279 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
1280 {
1281         struct vm_area_struct *vma;
1282         struct mm_struct *mm;
1283
1284         if (addr + len < addr)
1285                 return 0;
1286
1287         mm = get_task_mm(tsk);
1288         if (!mm)
1289                 return 0;
1290
1291         down_read(&mm->mmap_sem);
1292
1293         /* the access must start within one of the target process's mappings */
1294         vma = find_vma(mm, addr);
1295         if (vma) {
1296                 /* don't overrun this mapping */
1297                 if (addr + len >= vma->vm_end)
1298                         len = vma->vm_end - addr;
1299
1300                 /* only read or write mappings where it is permitted */
1301                 if (write && vma->vm_flags & VM_MAYWRITE)
1302                         len -= copy_to_user((void *) addr, buf, len);
1303                 else if (!write && vma->vm_flags & VM_MAYREAD)
1304                         len -= copy_from_user(buf, (void *) addr, len);
1305                 else
1306                         len = 0;
1307         } else {
1308                 len = 0;
1309         }
1310
1311         up_read(&mm->mmap_sem);
1312         mmput(mm);
1313         return len;
1314 }