xref: /openbmc/linux/mm/nommu.c (revision d5cb9783536a41df9f9cba5b0a1d78047ed787f7)
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(sysctl_max_map_count);
48 EXPORT_SYMBOL(sysctl_overcommit_memory);
49 EXPORT_SYMBOL(sysctl_overcommit_ratio);
50 EXPORT_SYMBOL(vm_committed_space);
51 EXPORT_SYMBOL(__vm_enough_memory);
52 
53 /* list of shareable VMAs */
54 struct rb_root nommu_vma_tree = RB_ROOT;
55 DECLARE_RWSEM(nommu_vma_sem);
56 
57 struct vm_operations_struct generic_file_vm_ops = {
58 };
59 
60 EXPORT_SYMBOL(vmalloc);
61 EXPORT_SYMBOL(vfree);
62 EXPORT_SYMBOL(vmalloc_to_page);
63 EXPORT_SYMBOL(vmalloc_32);
64 
65 /*
66  * Handle all mappings that got truncated by a "truncate()"
67  * system call.
68  *
69  * NOTE! We have to be ready to update the memory sharing
70  * between the file and the memory map for a potential last
71  * incomplete page.  Ugly, but necessary.
72  */
73 int vmtruncate(struct inode *inode, loff_t offset)
74 {
75 	struct address_space *mapping = inode->i_mapping;
76 	unsigned long limit;
77 
78 	if (inode->i_size < offset)
79 		goto do_expand;
80 	i_size_write(inode, offset);
81 
82 	truncate_inode_pages(mapping, offset);
83 	goto out_truncate;
84 
85 do_expand:
86 	limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
87 	if (limit != RLIM_INFINITY && offset > limit)
88 		goto out_sig;
89 	if (offset > inode->i_sb->s_maxbytes)
90 		goto out;
91 	i_size_write(inode, offset);
92 
93 out_truncate:
94 	if (inode->i_op && inode->i_op->truncate)
95 		inode->i_op->truncate(inode);
96 	return 0;
97 out_sig:
98 	send_sig(SIGXFSZ, current, 0);
99 out:
100 	return -EFBIG;
101 }
102 
103 EXPORT_SYMBOL(vmtruncate);
104 
105 /*
106  * Return the total memory allocated for this pointer, not
107  * just what the caller asked for.
108  *
109  * Doesn't have to be accurate, i.e. may have races.
110  */
111 unsigned int kobjsize(const void *objp)
112 {
113 	struct page *page;
114 
115 	if (!objp || !((page = virt_to_page(objp))))
116 		return 0;
117 
118 	if (PageSlab(page))
119 		return ksize(objp);
120 
121 	BUG_ON(page->index < 0);
122 	BUG_ON(page->index >= MAX_ORDER);
123 
124 	return (PAGE_SIZE << page->index);
125 }
126 
127 /*
128  * The nommu dodgy version :-)
129  */
130 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
131 	unsigned long start, int len, int write, int force,
132 	struct page **pages, struct vm_area_struct **vmas)
133 {
134 	int i;
135 	static struct vm_area_struct dummy_vma;
136 
137 	for (i = 0; i < len; i++) {
138 		if (pages) {
139 			pages[i] = virt_to_page(start);
140 			if (pages[i])
141 				page_cache_get(pages[i]);
142 		}
143 		if (vmas)
144 			vmas[i] = &dummy_vma;
145 		start += PAGE_SIZE;
146 	}
147 	return(i);
148 }
149 
150 EXPORT_SYMBOL(get_user_pages);
151 
152 DEFINE_RWLOCK(vmlist_lock);
153 struct vm_struct *vmlist;
154 
155 void vfree(void *addr)
156 {
157 	kfree(addr);
158 }
159 
160 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
161 {
162 	/*
163 	 * kmalloc doesn't like __GFP_HIGHMEM for some reason
164 	 */
165 	return kmalloc(size, gfp_mask & ~__GFP_HIGHMEM);
166 }
167 
168 struct page * vmalloc_to_page(void *addr)
169 {
170 	return virt_to_page(addr);
171 }
172 
173 unsigned long vmalloc_to_pfn(void *addr)
174 {
175 	return page_to_pfn(virt_to_page(addr));
176 }
177 
178 
179 long vread(char *buf, char *addr, unsigned long count)
180 {
181 	memcpy(buf, addr, count);
182 	return count;
183 }
184 
185 long vwrite(char *buf, char *addr, unsigned long count)
186 {
187 	/* Don't allow overflow */
188 	if ((unsigned long) addr + count < count)
189 		count = -(unsigned long) addr;
190 
191 	memcpy(addr, buf, count);
192 	return(count);
193 }
194 
195 /*
196  *	vmalloc  -  allocate virtually continguos memory
197  *
198  *	@size:		allocation size
199  *
200  *	Allocate enough pages to cover @size from the page level
201  *	allocator and map them into continguos kernel virtual space.
202  *
203  *	For tight cotrol over page level allocator and protection flags
204  *	use __vmalloc() instead.
205  */
206 void *vmalloc(unsigned long size)
207 {
208        return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
209 }
210 
211 /*
212  *	vmalloc_32  -  allocate virtually continguos memory (32bit addressable)
213  *
214  *	@size:		allocation size
215  *
216  *	Allocate enough 32bit PA addressable pages to cover @size from the
217  *	page level allocator and map them into continguos kernel virtual space.
218  */
219 void *vmalloc_32(unsigned long size)
220 {
221 	return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
222 }
223 
224 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
225 {
226 	BUG();
227 	return NULL;
228 }
229 
230 void vunmap(void *addr)
231 {
232 	BUG();
233 }
234 
235 /*
236  *  sys_brk() for the most part doesn't need the global kernel
237  *  lock, except when an application is doing something nasty
238  *  like trying to un-brk an area that has already been mapped
239  *  to a regular file.  in this case, the unmapping will need
240  *  to invoke file system routines that need the global lock.
241  */
242 asmlinkage unsigned long sys_brk(unsigned long brk)
243 {
244 	struct mm_struct *mm = current->mm;
245 
246 	if (brk < mm->start_brk || brk > mm->context.end_brk)
247 		return mm->brk;
248 
249 	if (mm->brk == brk)
250 		return mm->brk;
251 
252 	/*
253 	 * Always allow shrinking brk
254 	 */
255 	if (brk <= mm->brk) {
256 		mm->brk = brk;
257 		return brk;
258 	}
259 
260 	/*
261 	 * Ok, looks good - let it rip.
262 	 */
263 	return mm->brk = brk;
264 }
265 
266 #ifdef DEBUG
267 static void show_process_blocks(void)
268 {
269 	struct vm_list_struct *vml;
270 
271 	printk("Process blocks %d:", current->pid);
272 
273 	for (vml = &current->mm->context.vmlist; vml; vml = vml->next) {
274 		printk(" %p: %p", vml, vml->vma);
275 		if (vml->vma)
276 			printk(" (%d @%lx #%d)",
277 			       kobjsize((void *) vml->vma->vm_start),
278 			       vml->vma->vm_start,
279 			       atomic_read(&vml->vma->vm_usage));
280 		printk(vml->next ? " ->" : ".\n");
281 	}
282 }
283 #endif /* DEBUG */
284 
285 static inline struct vm_area_struct *find_nommu_vma(unsigned long start)
286 {
287 	struct vm_area_struct *vma;
288 	struct rb_node *n = nommu_vma_tree.rb_node;
289 
290 	while (n) {
291 		vma = rb_entry(n, struct vm_area_struct, vm_rb);
292 
293 		if (start < vma->vm_start)
294 			n = n->rb_left;
295 		else if (start > vma->vm_start)
296 			n = n->rb_right;
297 		else
298 			return vma;
299 	}
300 
301 	return NULL;
302 }
303 
304 static void add_nommu_vma(struct vm_area_struct *vma)
305 {
306 	struct vm_area_struct *pvma;
307 	struct address_space *mapping;
308 	struct rb_node **p = &nommu_vma_tree.rb_node;
309 	struct rb_node *parent = NULL;
310 
311 	/* add the VMA to the mapping */
312 	if (vma->vm_file) {
313 		mapping = vma->vm_file->f_mapping;
314 
315 		flush_dcache_mmap_lock(mapping);
316 		vma_prio_tree_insert(vma, &mapping->i_mmap);
317 		flush_dcache_mmap_unlock(mapping);
318 	}
319 
320 	/* add the VMA to the master list */
321 	while (*p) {
322 		parent = *p;
323 		pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
324 
325 		if (vma->vm_start < pvma->vm_start) {
326 			p = &(*p)->rb_left;
327 		}
328 		else if (vma->vm_start > pvma->vm_start) {
329 			p = &(*p)->rb_right;
330 		}
331 		else {
332 			/* mappings are at the same address - this can only
333 			 * happen for shared-mem chardevs and shared file
334 			 * mappings backed by ramfs/tmpfs */
335 			BUG_ON(!(pvma->vm_flags & VM_SHARED));
336 
337 			if (vma < pvma)
338 				p = &(*p)->rb_left;
339 			else if (vma > pvma)
340 				p = &(*p)->rb_right;
341 			else
342 				BUG();
343 		}
344 	}
345 
346 	rb_link_node(&vma->vm_rb, parent, p);
347 	rb_insert_color(&vma->vm_rb, &nommu_vma_tree);
348 }
349 
350 static void delete_nommu_vma(struct vm_area_struct *vma)
351 {
352 	struct address_space *mapping;
353 
354 	/* remove the VMA from the mapping */
355 	if (vma->vm_file) {
356 		mapping = vma->vm_file->f_mapping;
357 
358 		flush_dcache_mmap_lock(mapping);
359 		vma_prio_tree_remove(vma, &mapping->i_mmap);
360 		flush_dcache_mmap_unlock(mapping);
361 	}
362 
363 	/* remove from the master list */
364 	rb_erase(&vma->vm_rb, &nommu_vma_tree);
365 }
366 
367 /*
368  * determine whether a mapping should be permitted and, if so, what sort of
369  * mapping we're capable of supporting
370  */
371 static int validate_mmap_request(struct file *file,
372 				 unsigned long addr,
373 				 unsigned long len,
374 				 unsigned long prot,
375 				 unsigned long flags,
376 				 unsigned long pgoff,
377 				 unsigned long *_capabilities)
378 {
379 	unsigned long capabilities;
380 	unsigned long reqprot = prot;
381 	int ret;
382 
383 	/* do the simple checks first */
384 	if (flags & MAP_FIXED || addr) {
385 		printk(KERN_DEBUG
386 		       "%d: Can't do fixed-address/overlay mmap of RAM\n",
387 		       current->pid);
388 		return -EINVAL;
389 	}
390 
391 	if ((flags & MAP_TYPE) != MAP_PRIVATE &&
392 	    (flags & MAP_TYPE) != MAP_SHARED)
393 		return -EINVAL;
394 
395 	if (PAGE_ALIGN(len) == 0)
396 		return addr;
397 
398 	if (len > TASK_SIZE)
399 		return -EINVAL;
400 
401 	/* offset overflow? */
402 	if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
403 		return -EINVAL;
404 
405 	if (file) {
406 		/* validate file mapping requests */
407 		struct address_space *mapping;
408 
409 		/* files must support mmap */
410 		if (!file->f_op || !file->f_op->mmap)
411 			return -ENODEV;
412 
413 		/* work out if what we've got could possibly be shared
414 		 * - we support chardevs that provide their own "memory"
415 		 * - we support files/blockdevs that are memory backed
416 		 */
417 		mapping = file->f_mapping;
418 		if (!mapping)
419 			mapping = file->f_dentry->d_inode->i_mapping;
420 
421 		capabilities = 0;
422 		if (mapping && mapping->backing_dev_info)
423 			capabilities = mapping->backing_dev_info->capabilities;
424 
425 		if (!capabilities) {
426 			/* no explicit capabilities set, so assume some
427 			 * defaults */
428 			switch (file->f_dentry->d_inode->i_mode & S_IFMT) {
429 			case S_IFREG:
430 			case S_IFBLK:
431 				capabilities = BDI_CAP_MAP_COPY;
432 				break;
433 
434 			case S_IFCHR:
435 				capabilities =
436 					BDI_CAP_MAP_DIRECT |
437 					BDI_CAP_READ_MAP |
438 					BDI_CAP_WRITE_MAP;
439 				break;
440 
441 			default:
442 				return -EINVAL;
443 			}
444 		}
445 
446 		/* eliminate any capabilities that we can't support on this
447 		 * device */
448 		if (!file->f_op->get_unmapped_area)
449 			capabilities &= ~BDI_CAP_MAP_DIRECT;
450 		if (!file->f_op->read)
451 			capabilities &= ~BDI_CAP_MAP_COPY;
452 
453 		if (flags & MAP_SHARED) {
454 			/* do checks for writing, appending and locking */
455 			if ((prot & PROT_WRITE) &&
456 			    !(file->f_mode & FMODE_WRITE))
457 				return -EACCES;
458 
459 			if (IS_APPEND(file->f_dentry->d_inode) &&
460 			    (file->f_mode & FMODE_WRITE))
461 				return -EACCES;
462 
463 			if (locks_verify_locked(file->f_dentry->d_inode))
464 				return -EAGAIN;
465 
466 			if (!(capabilities & BDI_CAP_MAP_DIRECT))
467 				return -ENODEV;
468 
469 			if (((prot & PROT_READ)  && !(capabilities & BDI_CAP_READ_MAP))  ||
470 			    ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
471 			    ((prot & PROT_EXEC)  && !(capabilities & BDI_CAP_EXEC_MAP))
472 			    ) {
473 				printk("MAP_SHARED not completely supported on !MMU\n");
474 				return -EINVAL;
475 			}
476 
477 			/* we mustn't privatise shared mappings */
478 			capabilities &= ~BDI_CAP_MAP_COPY;
479 		}
480 		else {
481 			/* we're going to read the file into private memory we
482 			 * allocate */
483 			if (!(capabilities & BDI_CAP_MAP_COPY))
484 				return -ENODEV;
485 
486 			/* we don't permit a private writable mapping to be
487 			 * shared with the backing device */
488 			if (prot & PROT_WRITE)
489 				capabilities &= ~BDI_CAP_MAP_DIRECT;
490 		}
491 
492 		/* handle executable mappings and implied executable
493 		 * mappings */
494 		if (file->f_vfsmnt->mnt_flags & MNT_NOEXEC) {
495 			if (prot & PROT_EXEC)
496 				return -EPERM;
497 		}
498 		else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
499 			/* handle implication of PROT_EXEC by PROT_READ */
500 			if (current->personality & READ_IMPLIES_EXEC) {
501 				if (capabilities & BDI_CAP_EXEC_MAP)
502 					prot |= PROT_EXEC;
503 			}
504 		}
505 		else if ((prot & PROT_READ) &&
506 			 (prot & PROT_EXEC) &&
507 			 !(capabilities & BDI_CAP_EXEC_MAP)
508 			 ) {
509 			/* backing file is not executable, try to copy */
510 			capabilities &= ~BDI_CAP_MAP_DIRECT;
511 		}
512 	}
513 	else {
514 		/* anonymous mappings are always memory backed and can be
515 		 * privately mapped
516 		 */
517 		capabilities = BDI_CAP_MAP_COPY;
518 
519 		/* handle PROT_EXEC implication by PROT_READ */
520 		if ((prot & PROT_READ) &&
521 		    (current->personality & READ_IMPLIES_EXEC))
522 			prot |= PROT_EXEC;
523 	}
524 
525 	/* allow the security API to have its say */
526 	ret = security_file_mmap(file, reqprot, prot, flags);
527 	if (ret < 0)
528 		return ret;
529 
530 	/* looks okay */
531 	*_capabilities = capabilities;
532 	return 0;
533 }
534 
535 /*
536  * we've determined that we can make the mapping, now translate what we
537  * now know into VMA flags
538  */
539 static unsigned long determine_vm_flags(struct file *file,
540 					unsigned long prot,
541 					unsigned long flags,
542 					unsigned long capabilities)
543 {
544 	unsigned long vm_flags;
545 
546 	vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
547 	vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
548 	/* vm_flags |= mm->def_flags; */
549 
550 	if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
551 		/* attempt to share read-only copies of mapped file chunks */
552 		if (file && !(prot & PROT_WRITE))
553 			vm_flags |= VM_MAYSHARE;
554 	}
555 	else {
556 		/* overlay a shareable mapping on the backing device or inode
557 		 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
558 		 * romfs/cramfs */
559 		if (flags & MAP_SHARED)
560 			vm_flags |= VM_MAYSHARE | VM_SHARED;
561 		else if ((((vm_flags & capabilities) ^ vm_flags) & BDI_CAP_VMFLAGS) == 0)
562 			vm_flags |= VM_MAYSHARE;
563 	}
564 
565 	/* refuse to let anyone share private mappings with this process if
566 	 * it's being traced - otherwise breakpoints set in it may interfere
567 	 * with another untraced process
568 	 */
569 	if ((flags & MAP_PRIVATE) && (current->ptrace & PT_PTRACED))
570 		vm_flags &= ~VM_MAYSHARE;
571 
572 	return vm_flags;
573 }
574 
575 /*
576  * set up a shared mapping on a file
577  */
578 static int do_mmap_shared_file(struct vm_area_struct *vma, unsigned long len)
579 {
580 	int ret;
581 
582 	ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
583 	if (ret != -ENOSYS)
584 		return ret;
585 
586 	/* getting an ENOSYS error indicates that direct mmap isn't
587 	 * possible (as opposed to tried but failed) so we'll fall
588 	 * through to making a private copy of the data and mapping
589 	 * that if we can */
590 	return -ENODEV;
591 }
592 
593 /*
594  * set up a private mapping or an anonymous shared mapping
595  */
596 static int do_mmap_private(struct vm_area_struct *vma, unsigned long len)
597 {
598 	void *base;
599 	int ret;
600 
601 	/* invoke the file's mapping function so that it can keep track of
602 	 * shared mappings on devices or memory
603 	 * - VM_MAYSHARE will be set if it may attempt to share
604 	 */
605 	if (vma->vm_file) {
606 		ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
607 		if (ret != -ENOSYS) {
608 			/* shouldn't return success if we're not sharing */
609 			BUG_ON(ret == 0 && !(vma->vm_flags & VM_MAYSHARE));
610 			return ret; /* success or a real error */
611 		}
612 
613 		/* getting an ENOSYS error indicates that direct mmap isn't
614 		 * possible (as opposed to tried but failed) so we'll try to
615 		 * make a private copy of the data and map that instead */
616 	}
617 
618 	/* allocate some memory to hold the mapping
619 	 * - note that this may not return a page-aligned address if the object
620 	 *   we're allocating is smaller than a page
621 	 */
622 	base = kmalloc(len, GFP_KERNEL);
623 	if (!base)
624 		goto enomem;
625 
626 	vma->vm_start = (unsigned long) base;
627 	vma->vm_end = vma->vm_start + len;
628 	vma->vm_flags |= VM_MAPPED_COPY;
629 
630 #ifdef WARN_ON_SLACK
631 	if (len + WARN_ON_SLACK <= kobjsize(result))
632 		printk("Allocation of %lu bytes from process %d has %lu bytes of slack\n",
633 		       len, current->pid, kobjsize(result) - len);
634 #endif
635 
636 	if (vma->vm_file) {
637 		/* read the contents of a file into the copy */
638 		mm_segment_t old_fs;
639 		loff_t fpos;
640 
641 		fpos = vma->vm_pgoff;
642 		fpos <<= PAGE_SHIFT;
643 
644 		old_fs = get_fs();
645 		set_fs(KERNEL_DS);
646 		ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos);
647 		set_fs(old_fs);
648 
649 		if (ret < 0)
650 			goto error_free;
651 
652 		/* clear the last little bit */
653 		if (ret < len)
654 			memset(base + ret, 0, len - ret);
655 
656 	} else {
657 		/* if it's an anonymous mapping, then just clear it */
658 		memset(base, 0, len);
659 	}
660 
661 	return 0;
662 
663 error_free:
664 	kfree(base);
665 	vma->vm_start = 0;
666 	return ret;
667 
668 enomem:
669 	printk("Allocation of length %lu from process %d failed\n",
670 	       len, current->pid);
671 	show_free_areas();
672 	return -ENOMEM;
673 }
674 
675 /*
676  * handle mapping creation for uClinux
677  */
678 unsigned long do_mmap_pgoff(struct file *file,
679 			    unsigned long addr,
680 			    unsigned long len,
681 			    unsigned long prot,
682 			    unsigned long flags,
683 			    unsigned long pgoff)
684 {
685 	struct vm_list_struct *vml = NULL;
686 	struct vm_area_struct *vma = NULL;
687 	struct rb_node *rb;
688 	unsigned long capabilities, vm_flags;
689 	void *result;
690 	int ret;
691 
692 	/* decide whether we should attempt the mapping, and if so what sort of
693 	 * mapping */
694 	ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
695 				    &capabilities);
696 	if (ret < 0)
697 		return ret;
698 
699 	/* we've determined that we can make the mapping, now translate what we
700 	 * now know into VMA flags */
701 	vm_flags = determine_vm_flags(file, prot, flags, capabilities);
702 
703 	/* we're going to need to record the mapping if it works */
704 	vml = kmalloc(sizeof(struct vm_list_struct), GFP_KERNEL);
705 	if (!vml)
706 		goto error_getting_vml;
707 	memset(vml, 0, sizeof(*vml));
708 
709 	down_write(&nommu_vma_sem);
710 
711 	/* if we want to share, we need to check for VMAs created by other
712 	 * mmap() calls that overlap with our proposed mapping
713 	 * - we can only share with an exact match on most regular files
714 	 * - shared mappings on character devices and memory backed files are
715 	 *   permitted to overlap inexactly as far as we are concerned for in
716 	 *   these cases, sharing is handled in the driver or filesystem rather
717 	 *   than here
718 	 */
719 	if (vm_flags & VM_MAYSHARE) {
720 		unsigned long pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
721 		unsigned long vmpglen;
722 
723 		for (rb = rb_first(&nommu_vma_tree); rb; rb = rb_next(rb)) {
724 			vma = rb_entry(rb, struct vm_area_struct, vm_rb);
725 
726 			if (!(vma->vm_flags & VM_MAYSHARE))
727 				continue;
728 
729 			/* search for overlapping mappings on the same file */
730 			if (vma->vm_file->f_dentry->d_inode != file->f_dentry->d_inode)
731 				continue;
732 
733 			if (vma->vm_pgoff >= pgoff + pglen)
734 				continue;
735 
736 			vmpglen = vma->vm_end - vma->vm_start + PAGE_SIZE - 1;
737 			vmpglen >>= PAGE_SHIFT;
738 			if (pgoff >= vma->vm_pgoff + vmpglen)
739 				continue;
740 
741 			/* handle inexactly overlapping matches between mappings */
742 			if (vma->vm_pgoff != pgoff || vmpglen != pglen) {
743 				if (!(capabilities & BDI_CAP_MAP_DIRECT))
744 					goto sharing_violation;
745 				continue;
746 			}
747 
748 			/* we've found a VMA we can share */
749 			atomic_inc(&vma->vm_usage);
750 
751 			vml->vma = vma;
752 			result = (void *) vma->vm_start;
753 			goto shared;
754 		}
755 
756 		vma = NULL;
757 
758 		/* obtain the address at which to make a shared mapping
759 		 * - this is the hook for quasi-memory character devices to
760 		 *   tell us the location of a shared mapping
761 		 */
762 		if (file && file->f_op->get_unmapped_area) {
763 			addr = file->f_op->get_unmapped_area(file, addr, len,
764 							     pgoff, flags);
765 			if (IS_ERR((void *) addr)) {
766 				ret = addr;
767 				if (ret != (unsigned long) -ENOSYS)
768 					goto error;
769 
770 				/* the driver refused to tell us where to site
771 				 * the mapping so we'll have to attempt to copy
772 				 * it */
773 				ret = (unsigned long) -ENODEV;
774 				if (!(capabilities & BDI_CAP_MAP_COPY))
775 					goto error;
776 
777 				capabilities &= ~BDI_CAP_MAP_DIRECT;
778 			}
779 		}
780 	}
781 
782 	/* we're going to need a VMA struct as well */
783 	vma = kmalloc(sizeof(struct vm_area_struct), GFP_KERNEL);
784 	if (!vma)
785 		goto error_getting_vma;
786 
787 	memset(vma, 0, sizeof(*vma));
788 	INIT_LIST_HEAD(&vma->anon_vma_node);
789 	atomic_set(&vma->vm_usage, 1);
790 	if (file)
791 		get_file(file);
792 	vma->vm_file	= file;
793 	vma->vm_flags	= vm_flags;
794 	vma->vm_start	= addr;
795 	vma->vm_end	= addr + len;
796 	vma->vm_pgoff	= pgoff;
797 
798 	vml->vma = vma;
799 
800 	/* set up the mapping */
801 	if (file && vma->vm_flags & VM_SHARED)
802 		ret = do_mmap_shared_file(vma, len);
803 	else
804 		ret = do_mmap_private(vma, len);
805 	if (ret < 0)
806 		goto error;
807 
808 	/* okay... we have a mapping; now we have to register it */
809 	result = (void *) vma->vm_start;
810 
811 	if (vma->vm_flags & VM_MAPPED_COPY) {
812 		realalloc += kobjsize(result);
813 		askedalloc += len;
814 	}
815 
816 	realalloc += kobjsize(vma);
817 	askedalloc += sizeof(*vma);
818 
819 	current->mm->total_vm += len >> PAGE_SHIFT;
820 
821 	add_nommu_vma(vma);
822 
823  shared:
824 	realalloc += kobjsize(vml);
825 	askedalloc += sizeof(*vml);
826 
827 	vml->next = current->mm->context.vmlist;
828 	current->mm->context.vmlist = vml;
829 
830 	up_write(&nommu_vma_sem);
831 
832 	if (prot & PROT_EXEC)
833 		flush_icache_range((unsigned long) result,
834 				   (unsigned long) result + len);
835 
836 #ifdef DEBUG
837 	printk("do_mmap:\n");
838 	show_process_blocks();
839 #endif
840 
841 	return (unsigned long) result;
842 
843  error:
844 	up_write(&nommu_vma_sem);
845 	kfree(vml);
846 	if (vma) {
847 		fput(vma->vm_file);
848 		kfree(vma);
849 	}
850 	return ret;
851 
852  sharing_violation:
853 	up_write(&nommu_vma_sem);
854 	printk("Attempt to share mismatched mappings\n");
855 	kfree(vml);
856 	return -EINVAL;
857 
858  error_getting_vma:
859 	up_write(&nommu_vma_sem);
860 	kfree(vml);
861 	printk("Allocation of vma for %lu byte allocation from process %d failed\n",
862 	       len, current->pid);
863 	show_free_areas();
864 	return -ENOMEM;
865 
866  error_getting_vml:
867 	printk("Allocation of vml for %lu byte allocation from process %d failed\n",
868 	       len, current->pid);
869 	show_free_areas();
870 	return -ENOMEM;
871 }
872 
873 /*
874  * handle mapping disposal for uClinux
875  */
876 static void put_vma(struct vm_area_struct *vma)
877 {
878 	if (vma) {
879 		down_write(&nommu_vma_sem);
880 
881 		if (atomic_dec_and_test(&vma->vm_usage)) {
882 			delete_nommu_vma(vma);
883 
884 			if (vma->vm_ops && vma->vm_ops->close)
885 				vma->vm_ops->close(vma);
886 
887 			/* IO memory and memory shared directly out of the pagecache from
888 			 * ramfs/tmpfs mustn't be released here */
889 			if (vma->vm_flags & VM_MAPPED_COPY) {
890 				realalloc -= kobjsize((void *) vma->vm_start);
891 				askedalloc -= vma->vm_end - vma->vm_start;
892 				kfree((void *) vma->vm_start);
893 			}
894 
895 			realalloc -= kobjsize(vma);
896 			askedalloc -= sizeof(*vma);
897 
898 			if (vma->vm_file)
899 				fput(vma->vm_file);
900 			kfree(vma);
901 		}
902 
903 		up_write(&nommu_vma_sem);
904 	}
905 }
906 
907 int do_munmap(struct mm_struct *mm, unsigned long addr, size_t len)
908 {
909 	struct vm_list_struct *vml, **parent;
910 	unsigned long end = addr + len;
911 
912 #ifdef DEBUG
913 	printk("do_munmap:\n");
914 #endif
915 
916 	for (parent = &mm->context.vmlist; *parent; parent = &(*parent)->next)
917 		if ((*parent)->vma->vm_start == addr &&
918 		    ((len == 0) || ((*parent)->vma->vm_end == end)))
919 			goto found;
920 
921 	printk("munmap of non-mmaped memory by process %d (%s): %p\n",
922 	       current->pid, current->comm, (void *) addr);
923 	return -EINVAL;
924 
925  found:
926 	vml = *parent;
927 
928 	put_vma(vml->vma);
929 
930 	*parent = vml->next;
931 	realalloc -= kobjsize(vml);
932 	askedalloc -= sizeof(*vml);
933 	kfree(vml);
934 
935 	update_hiwater_vm(mm);
936 	mm->total_vm -= len >> PAGE_SHIFT;
937 
938 #ifdef DEBUG
939 	show_process_blocks();
940 #endif
941 
942 	return 0;
943 }
944 
945 /* Release all mmaps. */
946 void exit_mmap(struct mm_struct * mm)
947 {
948 	struct vm_list_struct *tmp;
949 
950 	if (mm) {
951 #ifdef DEBUG
952 		printk("Exit_mmap:\n");
953 #endif
954 
955 		mm->total_vm = 0;
956 
957 		while ((tmp = mm->context.vmlist)) {
958 			mm->context.vmlist = tmp->next;
959 			put_vma(tmp->vma);
960 
961 			realalloc -= kobjsize(tmp);
962 			askedalloc -= sizeof(*tmp);
963 			kfree(tmp);
964 		}
965 
966 #ifdef DEBUG
967 		show_process_blocks();
968 #endif
969 	}
970 }
971 
972 asmlinkage long sys_munmap(unsigned long addr, size_t len)
973 {
974 	int ret;
975 	struct mm_struct *mm = current->mm;
976 
977 	down_write(&mm->mmap_sem);
978 	ret = do_munmap(mm, addr, len);
979 	up_write(&mm->mmap_sem);
980 	return ret;
981 }
982 
983 unsigned long do_brk(unsigned long addr, unsigned long len)
984 {
985 	return -ENOMEM;
986 }
987 
988 /*
989  * Expand (or shrink) an existing mapping, potentially moving it at the
990  * same time (controlled by the MREMAP_MAYMOVE flag and available VM space)
991  *
992  * MREMAP_FIXED option added 5-Dec-1999 by Benjamin LaHaise
993  * This option implies MREMAP_MAYMOVE.
994  *
995  * on uClinux, we only permit changing a mapping's size, and only as long as it stays within the
996  * hole allocated by the kmalloc() call in do_mmap_pgoff() and the block is not shareable
997  */
998 unsigned long do_mremap(unsigned long addr,
999 			unsigned long old_len, unsigned long new_len,
1000 			unsigned long flags, unsigned long new_addr)
1001 {
1002 	struct vm_list_struct *vml = NULL;
1003 
1004 	/* insanity checks first */
1005 	if (new_len == 0)
1006 		return (unsigned long) -EINVAL;
1007 
1008 	if (flags & MREMAP_FIXED && new_addr != addr)
1009 		return (unsigned long) -EINVAL;
1010 
1011 	for (vml = current->mm->context.vmlist; vml; vml = vml->next)
1012 		if (vml->vma->vm_start == addr)
1013 			goto found;
1014 
1015 	return (unsigned long) -EINVAL;
1016 
1017  found:
1018 	if (vml->vma->vm_end != vml->vma->vm_start + old_len)
1019 		return (unsigned long) -EFAULT;
1020 
1021 	if (vml->vma->vm_flags & VM_MAYSHARE)
1022 		return (unsigned long) -EPERM;
1023 
1024 	if (new_len > kobjsize((void *) addr))
1025 		return (unsigned long) -ENOMEM;
1026 
1027 	/* all checks complete - do it */
1028 	vml->vma->vm_end = vml->vma->vm_start + new_len;
1029 
1030 	askedalloc -= old_len;
1031 	askedalloc += new_len;
1032 
1033 	return vml->vma->vm_start;
1034 }
1035 
1036 /*
1037  * Look up the first VMA which satisfies  addr < vm_end,  NULL if none
1038  */
1039 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
1040 {
1041 	struct vm_list_struct *vml;
1042 
1043 	for (vml = mm->context.vmlist; vml; vml = vml->next)
1044 		if (addr >= vml->vma->vm_start && addr < vml->vma->vm_end)
1045 			return vml->vma;
1046 
1047 	return NULL;
1048 }
1049 
1050 EXPORT_SYMBOL(find_vma);
1051 
1052 struct page *follow_page(struct mm_struct *mm, unsigned long address,
1053 			unsigned int foll_flags)
1054 {
1055 	return NULL;
1056 }
1057 
1058 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
1059 {
1060 	return NULL;
1061 }
1062 
1063 int remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
1064 		unsigned long to, unsigned long size, pgprot_t prot)
1065 {
1066 	vma->vm_start = vma->vm_pgoff << PAGE_SHIFT;
1067 	return 0;
1068 }
1069 
1070 void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1071 {
1072 }
1073 
1074 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1075 	unsigned long len, unsigned long pgoff, unsigned long flags)
1076 {
1077 	return -ENOMEM;
1078 }
1079 
1080 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1081 {
1082 }
1083 
1084 void unmap_mapping_range(struct address_space *mapping,
1085 			 loff_t const holebegin, loff_t const holelen,
1086 			 int even_cows)
1087 {
1088 }
1089 
1090 /*
1091  * Check that a process has enough memory to allocate a new virtual
1092  * mapping. 0 means there is enough memory for the allocation to
1093  * succeed and -ENOMEM implies there is not.
1094  *
1095  * We currently support three overcommit policies, which are set via the
1096  * vm.overcommit_memory sysctl.  See Documentation/vm/overcommit-accounting
1097  *
1098  * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1099  * Additional code 2002 Jul 20 by Robert Love.
1100  *
1101  * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1102  *
1103  * Note this is a helper function intended to be used by LSMs which
1104  * wish to use this logic.
1105  */
1106 int __vm_enough_memory(long pages, int cap_sys_admin)
1107 {
1108 	unsigned long free, allowed;
1109 
1110 	vm_acct_memory(pages);
1111 
1112 	/*
1113 	 * Sometimes we want to use more memory than we have
1114 	 */
1115 	if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1116 		return 0;
1117 
1118 	if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1119 		unsigned long n;
1120 
1121 		free = get_page_cache_size();
1122 		free += nr_swap_pages;
1123 
1124 		/*
1125 		 * Any slabs which are created with the
1126 		 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1127 		 * which are reclaimable, under pressure.  The dentry
1128 		 * cache and most inode caches should fall into this
1129 		 */
1130 		free += atomic_read(&slab_reclaim_pages);
1131 
1132 		/*
1133 		 * Leave the last 3% for root
1134 		 */
1135 		if (!cap_sys_admin)
1136 			free -= free / 32;
1137 
1138 		if (free > pages)
1139 			return 0;
1140 
1141 		/*
1142 		 * nr_free_pages() is very expensive on large systems,
1143 		 * only call if we're about to fail.
1144 		 */
1145 		n = nr_free_pages();
1146 		if (!cap_sys_admin)
1147 			n -= n / 32;
1148 		free += n;
1149 
1150 		if (free > pages)
1151 			return 0;
1152 		vm_unacct_memory(pages);
1153 		return -ENOMEM;
1154 	}
1155 
1156 	allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1157 	/*
1158 	 * Leave the last 3% for root
1159 	 */
1160 	if (!cap_sys_admin)
1161 		allowed -= allowed / 32;
1162 	allowed += total_swap_pages;
1163 
1164 	/* Don't let a single process grow too big:
1165 	   leave 3% of the size of this process for other processes */
1166 	allowed -= current->mm->total_vm / 32;
1167 
1168 	/*
1169 	 * cast `allowed' as a signed long because vm_committed_space
1170 	 * sometimes has a negative value
1171 	 */
1172 	if (atomic_read(&vm_committed_space) < (long)allowed)
1173 		return 0;
1174 
1175 	vm_unacct_memory(pages);
1176 
1177 	return -ENOMEM;
1178 }
1179 
1180 int in_gate_area_no_task(unsigned long addr)
1181 {
1182 	return 0;
1183 }
1184