xref: /linux/drivers/char/mem.c (revision 5e8d780d745c1619aba81fe7166c5a4b5cad2b84)
1 /*
2  *  linux/drivers/char/mem.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  *
6  *  Added devfs support.
7  *    Jan-11-1998, C. Scott Ananian <cananian@alumni.princeton.edu>
8  *  Shared /dev/zero mmaping support, Feb 2000, Kanoj Sarcar <kanoj@sgi.com>
9  */
10 
11 #include <linux/config.h>
12 #include <linux/mm.h>
13 #include <linux/miscdevice.h>
14 #include <linux/slab.h>
15 #include <linux/vmalloc.h>
16 #include <linux/mman.h>
17 #include <linux/random.h>
18 #include <linux/init.h>
19 #include <linux/raw.h>
20 #include <linux/tty.h>
21 #include <linux/capability.h>
22 #include <linux/smp_lock.h>
23 #include <linux/ptrace.h>
24 #include <linux/device.h>
25 #include <linux/highmem.h>
26 #include <linux/crash_dump.h>
27 #include <linux/backing-dev.h>
28 #include <linux/bootmem.h>
29 #include <linux/pipe_fs_i.h>
30 
31 #include <asm/uaccess.h>
32 #include <asm/io.h>
33 
34 #ifdef CONFIG_IA64
35 # include <linux/efi.h>
36 #endif
37 
38 /*
39  * Architectures vary in how they handle caching for addresses
40  * outside of main memory.
41  *
42  */
43 static inline int uncached_access(struct file *file, unsigned long addr)
44 {
45 #if defined(__i386__)
46 	/*
47 	 * On the PPro and successors, the MTRRs are used to set
48 	 * memory types for physical addresses outside main memory,
49 	 * so blindly setting PCD or PWT on those pages is wrong.
50 	 * For Pentiums and earlier, the surround logic should disable
51 	 * caching for the high addresses through the KEN pin, but
52 	 * we maintain the tradition of paranoia in this code.
53 	 */
54 	if (file->f_flags & O_SYNC)
55 		return 1;
56  	return !( test_bit(X86_FEATURE_MTRR, boot_cpu_data.x86_capability) ||
57 		  test_bit(X86_FEATURE_K6_MTRR, boot_cpu_data.x86_capability) ||
58 		  test_bit(X86_FEATURE_CYRIX_ARR, boot_cpu_data.x86_capability) ||
59 		  test_bit(X86_FEATURE_CENTAUR_MCR, boot_cpu_data.x86_capability) )
60 	  && addr >= __pa(high_memory);
61 #elif defined(__x86_64__)
62 	/*
63 	 * This is broken because it can generate memory type aliases,
64 	 * which can cause cache corruptions
65 	 * But it is only available for root and we have to be bug-to-bug
66 	 * compatible with i386.
67 	 */
68 	if (file->f_flags & O_SYNC)
69 		return 1;
70 	/* same behaviour as i386. PAT always set to cached and MTRRs control the
71 	   caching behaviour.
72 	   Hopefully a full PAT implementation will fix that soon. */
73 	return 0;
74 #elif defined(CONFIG_IA64)
75 	/*
76 	 * On ia64, we ignore O_SYNC because we cannot tolerate memory attribute aliases.
77 	 */
78 	return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
79 #else
80 	/*
81 	 * Accessing memory above the top the kernel knows about or through a file pointer
82 	 * that was marked O_SYNC will be done non-cached.
83 	 */
84 	if (file->f_flags & O_SYNC)
85 		return 1;
86 	return addr >= __pa(high_memory);
87 #endif
88 }
89 
90 #ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
91 static inline int valid_phys_addr_range(unsigned long addr, size_t count)
92 {
93 	if (addr + count > __pa(high_memory))
94 		return 0;
95 
96 	return 1;
97 }
98 
99 static inline int valid_mmap_phys_addr_range(unsigned long addr, size_t size)
100 {
101 	return 1;
102 }
103 #endif
104 
105 /*
106  * This funcion reads the *physical* memory. The f_pos points directly to the
107  * memory location.
108  */
109 static ssize_t read_mem(struct file * file, char __user * buf,
110 			size_t count, loff_t *ppos)
111 {
112 	unsigned long p = *ppos;
113 	ssize_t read, sz;
114 	char *ptr;
115 
116 	if (!valid_phys_addr_range(p, count))
117 		return -EFAULT;
118 	read = 0;
119 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
120 	/* we don't have page 0 mapped on sparc and m68k.. */
121 	if (p < PAGE_SIZE) {
122 		sz = PAGE_SIZE - p;
123 		if (sz > count)
124 			sz = count;
125 		if (sz > 0) {
126 			if (clear_user(buf, sz))
127 				return -EFAULT;
128 			buf += sz;
129 			p += sz;
130 			count -= sz;
131 			read += sz;
132 		}
133 	}
134 #endif
135 
136 	while (count > 0) {
137 		/*
138 		 * Handle first page in case it's not aligned
139 		 */
140 		if (-p & (PAGE_SIZE - 1))
141 			sz = -p & (PAGE_SIZE - 1);
142 		else
143 			sz = PAGE_SIZE;
144 
145 		sz = min_t(unsigned long, sz, count);
146 
147 		/*
148 		 * On ia64 if a page has been mapped somewhere as
149 		 * uncached, then it must also be accessed uncached
150 		 * by the kernel or data corruption may occur
151 		 */
152 		ptr = xlate_dev_mem_ptr(p);
153 
154 		if (copy_to_user(buf, ptr, sz))
155 			return -EFAULT;
156 		buf += sz;
157 		p += sz;
158 		count -= sz;
159 		read += sz;
160 	}
161 
162 	*ppos += read;
163 	return read;
164 }
165 
166 static ssize_t write_mem(struct file * file, const char __user * buf,
167 			 size_t count, loff_t *ppos)
168 {
169 	unsigned long p = *ppos;
170 	ssize_t written, sz;
171 	unsigned long copied;
172 	void *ptr;
173 
174 	if (!valid_phys_addr_range(p, count))
175 		return -EFAULT;
176 
177 	written = 0;
178 
179 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
180 	/* we don't have page 0 mapped on sparc and m68k.. */
181 	if (p < PAGE_SIZE) {
182 		unsigned long sz = PAGE_SIZE - p;
183 		if (sz > count)
184 			sz = count;
185 		/* Hmm. Do something? */
186 		buf += sz;
187 		p += sz;
188 		count -= sz;
189 		written += sz;
190 	}
191 #endif
192 
193 	while (count > 0) {
194 		/*
195 		 * Handle first page in case it's not aligned
196 		 */
197 		if (-p & (PAGE_SIZE - 1))
198 			sz = -p & (PAGE_SIZE - 1);
199 		else
200 			sz = PAGE_SIZE;
201 
202 		sz = min_t(unsigned long, sz, count);
203 
204 		/*
205 		 * On ia64 if a page has been mapped somewhere as
206 		 * uncached, then it must also be accessed uncached
207 		 * by the kernel or data corruption may occur
208 		 */
209 		ptr = xlate_dev_mem_ptr(p);
210 
211 		copied = copy_from_user(ptr, buf, sz);
212 		if (copied) {
213 			written += sz - copied;
214 			if (written)
215 				break;
216 			return -EFAULT;
217 		}
218 		buf += sz;
219 		p += sz;
220 		count -= sz;
221 		written += sz;
222 	}
223 
224 	*ppos += written;
225 	return written;
226 }
227 
228 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT
229 static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
230 				     unsigned long size, pgprot_t vma_prot)
231 {
232 #ifdef pgprot_noncached
233 	unsigned long offset = pfn << PAGE_SHIFT;
234 
235 	if (uncached_access(file, offset))
236 		return pgprot_noncached(vma_prot);
237 #endif
238 	return vma_prot;
239 }
240 #endif
241 
242 static int mmap_mem(struct file * file, struct vm_area_struct * vma)
243 {
244 	size_t size = vma->vm_end - vma->vm_start;
245 
246 	if (!valid_mmap_phys_addr_range(vma->vm_pgoff << PAGE_SHIFT, size))
247 		return -EINVAL;
248 
249 	vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
250 						 size,
251 						 vma->vm_page_prot);
252 
253 	/* Remap-pfn-range will mark the range VM_IO and VM_RESERVED */
254 	if (remap_pfn_range(vma,
255 			    vma->vm_start,
256 			    vma->vm_pgoff,
257 			    size,
258 			    vma->vm_page_prot))
259 		return -EAGAIN;
260 	return 0;
261 }
262 
263 static int mmap_kmem(struct file * file, struct vm_area_struct * vma)
264 {
265 	unsigned long pfn;
266 
267 	/* Turn a kernel-virtual address into a physical page frame */
268 	pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
269 
270 	/*
271 	 * RED-PEN: on some architectures there is more mapped memory
272 	 * than available in mem_map which pfn_valid checks
273 	 * for. Perhaps should add a new macro here.
274 	 *
275 	 * RED-PEN: vmalloc is not supported right now.
276 	 */
277 	if (!pfn_valid(pfn))
278 		return -EIO;
279 
280 	vma->vm_pgoff = pfn;
281 	return mmap_mem(file, vma);
282 }
283 
284 #ifdef CONFIG_CRASH_DUMP
285 /*
286  * Read memory corresponding to the old kernel.
287  */
288 static ssize_t read_oldmem(struct file *file, char __user *buf,
289 				size_t count, loff_t *ppos)
290 {
291 	unsigned long pfn, offset;
292 	size_t read = 0, csize;
293 	int rc = 0;
294 
295 	while (count) {
296 		pfn = *ppos / PAGE_SIZE;
297 		if (pfn > saved_max_pfn)
298 			return read;
299 
300 		offset = (unsigned long)(*ppos % PAGE_SIZE);
301 		if (count > PAGE_SIZE - offset)
302 			csize = PAGE_SIZE - offset;
303 		else
304 			csize = count;
305 
306 		rc = copy_oldmem_page(pfn, buf, csize, offset, 1);
307 		if (rc < 0)
308 			return rc;
309 		buf += csize;
310 		*ppos += csize;
311 		read += csize;
312 		count -= csize;
313 	}
314 	return read;
315 }
316 #endif
317 
318 extern long vread(char *buf, char *addr, unsigned long count);
319 extern long vwrite(char *buf, char *addr, unsigned long count);
320 
321 /*
322  * This function reads the *virtual* memory as seen by the kernel.
323  */
324 static ssize_t read_kmem(struct file *file, char __user *buf,
325 			 size_t count, loff_t *ppos)
326 {
327 	unsigned long p = *ppos;
328 	ssize_t low_count, read, sz;
329 	char * kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
330 
331 	read = 0;
332 	if (p < (unsigned long) high_memory) {
333 		low_count = count;
334 		if (count > (unsigned long) high_memory - p)
335 			low_count = (unsigned long) high_memory - p;
336 
337 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
338 		/* we don't have page 0 mapped on sparc and m68k.. */
339 		if (p < PAGE_SIZE && low_count > 0) {
340 			size_t tmp = PAGE_SIZE - p;
341 			if (tmp > low_count) tmp = low_count;
342 			if (clear_user(buf, tmp))
343 				return -EFAULT;
344 			buf += tmp;
345 			p += tmp;
346 			read += tmp;
347 			low_count -= tmp;
348 			count -= tmp;
349 		}
350 #endif
351 		while (low_count > 0) {
352 			/*
353 			 * Handle first page in case it's not aligned
354 			 */
355 			if (-p & (PAGE_SIZE - 1))
356 				sz = -p & (PAGE_SIZE - 1);
357 			else
358 				sz = PAGE_SIZE;
359 
360 			sz = min_t(unsigned long, sz, low_count);
361 
362 			/*
363 			 * On ia64 if a page has been mapped somewhere as
364 			 * uncached, then it must also be accessed uncached
365 			 * by the kernel or data corruption may occur
366 			 */
367 			kbuf = xlate_dev_kmem_ptr((char *)p);
368 
369 			if (copy_to_user(buf, kbuf, sz))
370 				return -EFAULT;
371 			buf += sz;
372 			p += sz;
373 			read += sz;
374 			low_count -= sz;
375 			count -= sz;
376 		}
377 	}
378 
379 	if (count > 0) {
380 		kbuf = (char *)__get_free_page(GFP_KERNEL);
381 		if (!kbuf)
382 			return -ENOMEM;
383 		while (count > 0) {
384 			int len = count;
385 
386 			if (len > PAGE_SIZE)
387 				len = PAGE_SIZE;
388 			len = vread(kbuf, (char *)p, len);
389 			if (!len)
390 				break;
391 			if (copy_to_user(buf, kbuf, len)) {
392 				free_page((unsigned long)kbuf);
393 				return -EFAULT;
394 			}
395 			count -= len;
396 			buf += len;
397 			read += len;
398 			p += len;
399 		}
400 		free_page((unsigned long)kbuf);
401 	}
402  	*ppos = p;
403  	return read;
404 }
405 
406 
407 static inline ssize_t
408 do_write_kmem(void *p, unsigned long realp, const char __user * buf,
409 	      size_t count, loff_t *ppos)
410 {
411 	ssize_t written, sz;
412 	unsigned long copied;
413 
414 	written = 0;
415 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
416 	/* we don't have page 0 mapped on sparc and m68k.. */
417 	if (realp < PAGE_SIZE) {
418 		unsigned long sz = PAGE_SIZE - realp;
419 		if (sz > count)
420 			sz = count;
421 		/* Hmm. Do something? */
422 		buf += sz;
423 		p += sz;
424 		realp += sz;
425 		count -= sz;
426 		written += sz;
427 	}
428 #endif
429 
430 	while (count > 0) {
431 		char *ptr;
432 		/*
433 		 * Handle first page in case it's not aligned
434 		 */
435 		if (-realp & (PAGE_SIZE - 1))
436 			sz = -realp & (PAGE_SIZE - 1);
437 		else
438 			sz = PAGE_SIZE;
439 
440 		sz = min_t(unsigned long, sz, count);
441 
442 		/*
443 		 * On ia64 if a page has been mapped somewhere as
444 		 * uncached, then it must also be accessed uncached
445 		 * by the kernel or data corruption may occur
446 		 */
447 		ptr = xlate_dev_kmem_ptr(p);
448 
449 		copied = copy_from_user(ptr, buf, sz);
450 		if (copied) {
451 			written += sz - copied;
452 			if (written)
453 				break;
454 			return -EFAULT;
455 		}
456 		buf += sz;
457 		p += sz;
458 		realp += sz;
459 		count -= sz;
460 		written += sz;
461 	}
462 
463 	*ppos += written;
464 	return written;
465 }
466 
467 
468 /*
469  * This function writes to the *virtual* memory as seen by the kernel.
470  */
471 static ssize_t write_kmem(struct file * file, const char __user * buf,
472 			  size_t count, loff_t *ppos)
473 {
474 	unsigned long p = *ppos;
475 	ssize_t wrote = 0;
476 	ssize_t virtr = 0;
477 	ssize_t written;
478 	char * kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
479 
480 	if (p < (unsigned long) high_memory) {
481 
482 		wrote = count;
483 		if (count > (unsigned long) high_memory - p)
484 			wrote = (unsigned long) high_memory - p;
485 
486 		written = do_write_kmem((void*)p, p, buf, wrote, ppos);
487 		if (written != wrote)
488 			return written;
489 		wrote = written;
490 		p += wrote;
491 		buf += wrote;
492 		count -= wrote;
493 	}
494 
495 	if (count > 0) {
496 		kbuf = (char *)__get_free_page(GFP_KERNEL);
497 		if (!kbuf)
498 			return wrote ? wrote : -ENOMEM;
499 		while (count > 0) {
500 			int len = count;
501 
502 			if (len > PAGE_SIZE)
503 				len = PAGE_SIZE;
504 			if (len) {
505 				written = copy_from_user(kbuf, buf, len);
506 				if (written) {
507 					if (wrote + virtr)
508 						break;
509 					free_page((unsigned long)kbuf);
510 					return -EFAULT;
511 				}
512 			}
513 			len = vwrite(kbuf, (char *)p, len);
514 			count -= len;
515 			buf += len;
516 			virtr += len;
517 			p += len;
518 		}
519 		free_page((unsigned long)kbuf);
520 	}
521 
522  	*ppos = p;
523  	return virtr + wrote;
524 }
525 
526 #if defined(CONFIG_ISA) || !defined(__mc68000__)
527 static ssize_t read_port(struct file * file, char __user * buf,
528 			 size_t count, loff_t *ppos)
529 {
530 	unsigned long i = *ppos;
531 	char __user *tmp = buf;
532 
533 	if (!access_ok(VERIFY_WRITE, buf, count))
534 		return -EFAULT;
535 	while (count-- > 0 && i < 65536) {
536 		if (__put_user(inb(i),tmp) < 0)
537 			return -EFAULT;
538 		i++;
539 		tmp++;
540 	}
541 	*ppos = i;
542 	return tmp-buf;
543 }
544 
545 static ssize_t write_port(struct file * file, const char __user * buf,
546 			  size_t count, loff_t *ppos)
547 {
548 	unsigned long i = *ppos;
549 	const char __user * tmp = buf;
550 
551 	if (!access_ok(VERIFY_READ,buf,count))
552 		return -EFAULT;
553 	while (count-- > 0 && i < 65536) {
554 		char c;
555 		if (__get_user(c, tmp)) {
556 			if (tmp > buf)
557 				break;
558 			return -EFAULT;
559 		}
560 		outb(c,i);
561 		i++;
562 		tmp++;
563 	}
564 	*ppos = i;
565 	return tmp-buf;
566 }
567 #endif
568 
569 static ssize_t read_null(struct file * file, char __user * buf,
570 			 size_t count, loff_t *ppos)
571 {
572 	return 0;
573 }
574 
575 static ssize_t write_null(struct file * file, const char __user * buf,
576 			  size_t count, loff_t *ppos)
577 {
578 	return count;
579 }
580 
581 static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
582 			struct splice_desc *sd)
583 {
584 	return sd->len;
585 }
586 
587 static ssize_t splice_write_null(struct pipe_inode_info *pipe,struct file *out,
588 				 loff_t *ppos, size_t len, unsigned int flags)
589 {
590 	return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
591 }
592 
593 #ifdef CONFIG_MMU
594 /*
595  * For fun, we are using the MMU for this.
596  */
597 static inline size_t read_zero_pagealigned(char __user * buf, size_t size)
598 {
599 	struct mm_struct *mm;
600 	struct vm_area_struct * vma;
601 	unsigned long addr=(unsigned long)buf;
602 
603 	mm = current->mm;
604 	/* Oops, this was forgotten before. -ben */
605 	down_read(&mm->mmap_sem);
606 
607 	/* For private mappings, just map in zero pages. */
608 	for (vma = find_vma(mm, addr); vma; vma = vma->vm_next) {
609 		unsigned long count;
610 
611 		if (vma->vm_start > addr || (vma->vm_flags & VM_WRITE) == 0)
612 			goto out_up;
613 		if (vma->vm_flags & (VM_SHARED | VM_HUGETLB))
614 			break;
615 		count = vma->vm_end - addr;
616 		if (count > size)
617 			count = size;
618 
619 		zap_page_range(vma, addr, count, NULL);
620         	zeromap_page_range(vma, addr, count, PAGE_COPY);
621 
622 		size -= count;
623 		buf += count;
624 		addr += count;
625 		if (size == 0)
626 			goto out_up;
627 	}
628 
629 	up_read(&mm->mmap_sem);
630 
631 	/* The shared case is hard. Let's do the conventional zeroing. */
632 	do {
633 		unsigned long unwritten = clear_user(buf, PAGE_SIZE);
634 		if (unwritten)
635 			return size + unwritten - PAGE_SIZE;
636 		cond_resched();
637 		buf += PAGE_SIZE;
638 		size -= PAGE_SIZE;
639 	} while (size);
640 
641 	return size;
642 out_up:
643 	up_read(&mm->mmap_sem);
644 	return size;
645 }
646 
647 static ssize_t read_zero(struct file * file, char __user * buf,
648 			 size_t count, loff_t *ppos)
649 {
650 	unsigned long left, unwritten, written = 0;
651 
652 	if (!count)
653 		return 0;
654 
655 	if (!access_ok(VERIFY_WRITE, buf, count))
656 		return -EFAULT;
657 
658 	left = count;
659 
660 	/* do we want to be clever? Arbitrary cut-off */
661 	if (count >= PAGE_SIZE*4) {
662 		unsigned long partial;
663 
664 		/* How much left of the page? */
665 		partial = (PAGE_SIZE-1) & -(unsigned long) buf;
666 		unwritten = clear_user(buf, partial);
667 		written = partial - unwritten;
668 		if (unwritten)
669 			goto out;
670 		left -= partial;
671 		buf += partial;
672 		unwritten = read_zero_pagealigned(buf, left & PAGE_MASK);
673 		written += (left & PAGE_MASK) - unwritten;
674 		if (unwritten)
675 			goto out;
676 		buf += left & PAGE_MASK;
677 		left &= ~PAGE_MASK;
678 	}
679 	unwritten = clear_user(buf, left);
680 	written += left - unwritten;
681 out:
682 	return written ? written : -EFAULT;
683 }
684 
685 static int mmap_zero(struct file * file, struct vm_area_struct * vma)
686 {
687 	if (vma->vm_flags & VM_SHARED)
688 		return shmem_zero_setup(vma);
689 	if (zeromap_page_range(vma, vma->vm_start, vma->vm_end - vma->vm_start, vma->vm_page_prot))
690 		return -EAGAIN;
691 	return 0;
692 }
693 #else /* CONFIG_MMU */
694 static ssize_t read_zero(struct file * file, char * buf,
695 			 size_t count, loff_t *ppos)
696 {
697 	size_t todo = count;
698 
699 	while (todo) {
700 		size_t chunk = todo;
701 
702 		if (chunk > 4096)
703 			chunk = 4096;	/* Just for latency reasons */
704 		if (clear_user(buf, chunk))
705 			return -EFAULT;
706 		buf += chunk;
707 		todo -= chunk;
708 		cond_resched();
709 	}
710 	return count;
711 }
712 
713 static int mmap_zero(struct file * file, struct vm_area_struct * vma)
714 {
715 	return -ENOSYS;
716 }
717 #endif /* CONFIG_MMU */
718 
719 static ssize_t write_full(struct file * file, const char __user * buf,
720 			  size_t count, loff_t *ppos)
721 {
722 	return -ENOSPC;
723 }
724 
725 /*
726  * Special lseek() function for /dev/null and /dev/zero.  Most notably, you
727  * can fopen() both devices with "a" now.  This was previously impossible.
728  * -- SRB.
729  */
730 
731 static loff_t null_lseek(struct file * file, loff_t offset, int orig)
732 {
733 	return file->f_pos = 0;
734 }
735 
736 /*
737  * The memory devices use the full 32/64 bits of the offset, and so we cannot
738  * check against negative addresses: they are ok. The return value is weird,
739  * though, in that case (0).
740  *
741  * also note that seeking relative to the "end of file" isn't supported:
742  * it has no meaning, so it returns -EINVAL.
743  */
744 static loff_t memory_lseek(struct file * file, loff_t offset, int orig)
745 {
746 	loff_t ret;
747 
748 	mutex_lock(&file->f_dentry->d_inode->i_mutex);
749 	switch (orig) {
750 		case 0:
751 			file->f_pos = offset;
752 			ret = file->f_pos;
753 			force_successful_syscall_return();
754 			break;
755 		case 1:
756 			file->f_pos += offset;
757 			ret = file->f_pos;
758 			force_successful_syscall_return();
759 			break;
760 		default:
761 			ret = -EINVAL;
762 	}
763 	mutex_unlock(&file->f_dentry->d_inode->i_mutex);
764 	return ret;
765 }
766 
767 static int open_port(struct inode * inode, struct file * filp)
768 {
769 	return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
770 }
771 
772 #define zero_lseek	null_lseek
773 #define full_lseek      null_lseek
774 #define write_zero	write_null
775 #define read_full       read_zero
776 #define open_mem	open_port
777 #define open_kmem	open_mem
778 #define open_oldmem	open_mem
779 
780 static struct file_operations mem_fops = {
781 	.llseek		= memory_lseek,
782 	.read		= read_mem,
783 	.write		= write_mem,
784 	.mmap		= mmap_mem,
785 	.open		= open_mem,
786 };
787 
788 static struct file_operations kmem_fops = {
789 	.llseek		= memory_lseek,
790 	.read		= read_kmem,
791 	.write		= write_kmem,
792 	.mmap		= mmap_kmem,
793 	.open		= open_kmem,
794 };
795 
796 static struct file_operations null_fops = {
797 	.llseek		= null_lseek,
798 	.read		= read_null,
799 	.write		= write_null,
800 	.splice_write	= splice_write_null,
801 };
802 
803 #if defined(CONFIG_ISA) || !defined(__mc68000__)
804 static struct file_operations port_fops = {
805 	.llseek		= memory_lseek,
806 	.read		= read_port,
807 	.write		= write_port,
808 	.open		= open_port,
809 };
810 #endif
811 
812 static struct file_operations zero_fops = {
813 	.llseek		= zero_lseek,
814 	.read		= read_zero,
815 	.write		= write_zero,
816 	.mmap		= mmap_zero,
817 };
818 
819 static struct backing_dev_info zero_bdi = {
820 	.capabilities	= BDI_CAP_MAP_COPY,
821 };
822 
823 static struct file_operations full_fops = {
824 	.llseek		= full_lseek,
825 	.read		= read_full,
826 	.write		= write_full,
827 };
828 
829 #ifdef CONFIG_CRASH_DUMP
830 static struct file_operations oldmem_fops = {
831 	.read	= read_oldmem,
832 	.open	= open_oldmem,
833 };
834 #endif
835 
836 static ssize_t kmsg_write(struct file * file, const char __user * buf,
837 			  size_t count, loff_t *ppos)
838 {
839 	char *tmp;
840 	ssize_t ret;
841 
842 	tmp = kmalloc(count + 1, GFP_KERNEL);
843 	if (tmp == NULL)
844 		return -ENOMEM;
845 	ret = -EFAULT;
846 	if (!copy_from_user(tmp, buf, count)) {
847 		tmp[count] = 0;
848 		ret = printk("%s", tmp);
849 		if (ret > count)
850 			/* printk can add a prefix */
851 			ret = count;
852 	}
853 	kfree(tmp);
854 	return ret;
855 }
856 
857 static struct file_operations kmsg_fops = {
858 	.write =	kmsg_write,
859 };
860 
861 static int memory_open(struct inode * inode, struct file * filp)
862 {
863 	switch (iminor(inode)) {
864 		case 1:
865 			filp->f_op = &mem_fops;
866 			break;
867 		case 2:
868 			filp->f_op = &kmem_fops;
869 			break;
870 		case 3:
871 			filp->f_op = &null_fops;
872 			break;
873 #if defined(CONFIG_ISA) || !defined(__mc68000__)
874 		case 4:
875 			filp->f_op = &port_fops;
876 			break;
877 #endif
878 		case 5:
879 			filp->f_mapping->backing_dev_info = &zero_bdi;
880 			filp->f_op = &zero_fops;
881 			break;
882 		case 7:
883 			filp->f_op = &full_fops;
884 			break;
885 		case 8:
886 			filp->f_op = &random_fops;
887 			break;
888 		case 9:
889 			filp->f_op = &urandom_fops;
890 			break;
891 		case 11:
892 			filp->f_op = &kmsg_fops;
893 			break;
894 #ifdef CONFIG_CRASH_DUMP
895 		case 12:
896 			filp->f_op = &oldmem_fops;
897 			break;
898 #endif
899 		default:
900 			return -ENXIO;
901 	}
902 	if (filp->f_op && filp->f_op->open)
903 		return filp->f_op->open(inode,filp);
904 	return 0;
905 }
906 
907 static struct file_operations memory_fops = {
908 	.open		= memory_open,	/* just a selector for the real open */
909 };
910 
911 static const struct {
912 	unsigned int		minor;
913 	char			*name;
914 	umode_t			mode;
915 	const struct file_operations	*fops;
916 } devlist[] = { /* list of minor devices */
917 	{1, "mem",     S_IRUSR | S_IWUSR | S_IRGRP, &mem_fops},
918 	{2, "kmem",    S_IRUSR | S_IWUSR | S_IRGRP, &kmem_fops},
919 	{3, "null",    S_IRUGO | S_IWUGO,           &null_fops},
920 #if defined(CONFIG_ISA) || !defined(__mc68000__)
921 	{4, "port",    S_IRUSR | S_IWUSR | S_IRGRP, &port_fops},
922 #endif
923 	{5, "zero",    S_IRUGO | S_IWUGO,           &zero_fops},
924 	{7, "full",    S_IRUGO | S_IWUGO,           &full_fops},
925 	{8, "random",  S_IRUGO | S_IWUSR,           &random_fops},
926 	{9, "urandom", S_IRUGO | S_IWUSR,           &urandom_fops},
927 	{11,"kmsg",    S_IRUGO | S_IWUSR,           &kmsg_fops},
928 #ifdef CONFIG_CRASH_DUMP
929 	{12,"oldmem",    S_IRUSR | S_IWUSR | S_IRGRP, &oldmem_fops},
930 #endif
931 };
932 
933 static struct class *mem_class;
934 
935 static int __init chr_dev_init(void)
936 {
937 	int i;
938 
939 	if (register_chrdev(MEM_MAJOR,"mem",&memory_fops))
940 		printk("unable to get major %d for memory devs\n", MEM_MAJOR);
941 
942 	mem_class = class_create(THIS_MODULE, "mem");
943 	for (i = 0; i < ARRAY_SIZE(devlist); i++)
944 		class_device_create(mem_class, NULL,
945 					MKDEV(MEM_MAJOR, devlist[i].minor),
946 					NULL, devlist[i].name);
947 
948 	return 0;
949 }
950 
951 fs_initcall(chr_dev_init);
952