• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 mmapping support, Feb 2000, Kanoj Sarcar <kanoj@sgi.com>
9  */
10 
11 #include <linux/mm.h>
12 #include <linux/miscdevice.h>
13 #include <linux/slab.h>
14 #include <linux/vmalloc.h>
15 #include <linux/mman.h>
16 #include <linux/random.h>
17 #include <linux/init.h>
18 #include <linux/raw.h>
19 #include <linux/tty.h>
20 #include <linux/capability.h>
21 #include <linux/ptrace.h>
22 #include <linux/device.h>
23 #include <linux/highmem.h>
24 #include <linux/backing-dev.h>
25 #include <linux/splice.h>
26 #include <linux/pfn.h>
27 #include <linux/export.h>
28 #include <linux/io.h>
29 #include <linux/aio.h>
30 
31 #include <asm/uaccess.h>
32 
33 #ifdef CONFIG_IA64
34 # include <linux/efi.h>
35 #endif
36 
37 #define DEVPORT_MINOR	4
38 
size_inside_page(unsigned long start,unsigned long size)39 static inline unsigned long size_inside_page(unsigned long start,
40 					     unsigned long size)
41 {
42 	unsigned long sz;
43 
44 	sz = PAGE_SIZE - (start & (PAGE_SIZE - 1));
45 
46 	return min(sz, size);
47 }
48 
49 #ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
valid_phys_addr_range(phys_addr_t addr,size_t count)50 static inline int valid_phys_addr_range(phys_addr_t addr, size_t count)
51 {
52 	return addr + count <= __pa(high_memory);
53 }
54 
valid_mmap_phys_addr_range(unsigned long pfn,size_t size)55 static inline int valid_mmap_phys_addr_range(unsigned long pfn, size_t size)
56 {
57 	return 1;
58 }
59 #endif
60 
61 #if defined(CONFIG_DEVMEM) || defined(CONFIG_DEVKMEM)
62 #ifdef CONFIG_STRICT_DEVMEM
page_is_allowed(unsigned long pfn)63 static inline int page_is_allowed(unsigned long pfn)
64 {
65 	return devmem_is_allowed(pfn);
66 }
range_is_allowed(unsigned long pfn,unsigned long size)67 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
68 {
69 	u64 from = ((u64)pfn) << PAGE_SHIFT;
70 	u64 to = from + size;
71 	u64 cursor = from;
72 
73 	while (cursor < to) {
74 		if (!devmem_is_allowed(pfn)) {
75 			printk(KERN_INFO
76 		"Program %s tried to access /dev/mem between %Lx->%Lx.\n",
77 				current->comm, from, to);
78 			return 0;
79 		}
80 		cursor += PAGE_SIZE;
81 		pfn++;
82 	}
83 	return 1;
84 }
85 #else
page_is_allowed(unsigned long pfn)86 static inline int page_is_allowed(unsigned long pfn)
87 {
88 	return 1;
89 }
range_is_allowed(unsigned long pfn,unsigned long size)90 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
91 {
92 	return 1;
93 }
94 #endif
95 #endif
96 
97 #ifdef CONFIG_DEVMEM
unxlate_dev_mem_ptr(unsigned long phys,void * addr)98 void __weak unxlate_dev_mem_ptr(unsigned long phys, void *addr)
99 {
100 }
101 
102 /*
103  * This funcion reads the *physical* memory. The f_pos points directly to the
104  * memory location.
105  */
read_mem(struct file * file,char __user * buf,size_t count,loff_t * ppos)106 static ssize_t read_mem(struct file *file, char __user *buf,
107 			size_t count, loff_t *ppos)
108 {
109 	phys_addr_t p = *ppos;
110 	ssize_t read, sz;
111 	char *ptr;
112 
113 	if (p != *ppos)
114 		return 0;
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 = size_inside_page(p, count);
123 		if (sz > 0) {
124 			if (clear_user(buf, sz))
125 				return -EFAULT;
126 			buf += sz;
127 			p += sz;
128 			count -= sz;
129 			read += sz;
130 		}
131 	}
132 #endif
133 
134 	while (count > 0) {
135 		unsigned long remaining;
136 		int allowed;
137 
138 		sz = size_inside_page(p, count);
139 
140 		allowed = page_is_allowed(p >> PAGE_SHIFT);
141 		if (!allowed)
142 			return -EPERM;
143 		if (allowed == 2) {
144 			/* Show zeros for restricted memory. */
145 			remaining = clear_user(buf, sz);
146 		} else {
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 			if (!ptr)
154 				return -EFAULT;
155 
156 			remaining = copy_to_user(buf, ptr, sz);
157 
158 			unxlate_dev_mem_ptr(p, ptr);
159 		}
160 
161 		if (remaining)
162 			return -EFAULT;
163 
164 		buf += sz;
165 		p += sz;
166 		count -= sz;
167 		read += sz;
168 	}
169 
170 	*ppos += read;
171 	return read;
172 }
173 
write_mem(struct file * file,const char __user * buf,size_t count,loff_t * ppos)174 static ssize_t write_mem(struct file *file, const char __user *buf,
175 			 size_t count, loff_t *ppos)
176 {
177 	phys_addr_t p = *ppos;
178 	ssize_t written, sz;
179 	unsigned long copied;
180 	void *ptr;
181 
182 	if (p != *ppos)
183 		return -EFBIG;
184 
185 	if (!valid_phys_addr_range(p, count))
186 		return -EFAULT;
187 
188 	written = 0;
189 
190 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
191 	/* we don't have page 0 mapped on sparc and m68k.. */
192 	if (p < PAGE_SIZE) {
193 		sz = size_inside_page(p, count);
194 		/* Hmm. Do something? */
195 		buf += sz;
196 		p += sz;
197 		count -= sz;
198 		written += sz;
199 	}
200 #endif
201 
202 	while (count > 0) {
203 		int allowed;
204 
205 		sz = size_inside_page(p, count);
206 
207 		allowed = page_is_allowed(p >> PAGE_SHIFT);
208 		if (!allowed)
209 			return -EPERM;
210 
211 		/* Skip actual writing when a page is marked as restricted. */
212 		if (allowed == 1) {
213 			/*
214 			 * On ia64 if a page has been mapped somewhere as
215 			 * uncached, then it must also be accessed uncached
216 			 * by the kernel or data corruption may occur.
217 			 */
218 			ptr = xlate_dev_mem_ptr(p);
219 			if (!ptr) {
220 				if (written)
221 					break;
222 				return -EFAULT;
223 			}
224 
225 			copied = copy_from_user(ptr, buf, sz);
226 			unxlate_dev_mem_ptr(p, ptr);
227 			if (copied) {
228 				written += sz - copied;
229 				if (written)
230 					break;
231 				return -EFAULT;
232 			}
233 		}
234 
235 		buf += sz;
236 		p += sz;
237 		count -= sz;
238 		written += sz;
239 	}
240 
241 	*ppos += written;
242 	return written;
243 }
244 #endif	/* CONFIG_DEVMEM */
245 
246 #if defined(CONFIG_DEVMEM) || defined(CONFIG_DEVKMEM)
247 
phys_mem_access_prot_allowed(struct file * file,unsigned long pfn,unsigned long size,pgprot_t * vma_prot)248 int __weak phys_mem_access_prot_allowed(struct file *file,
249 	unsigned long pfn, unsigned long size, pgprot_t *vma_prot)
250 {
251 	return 1;
252 }
253 
254 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT
255 
256 /*
257  * Architectures vary in how they handle caching for addresses
258  * outside of main memory.
259  *
260  */
261 #ifdef pgprot_noncached
uncached_access(struct file * file,phys_addr_t addr)262 static int uncached_access(struct file *file, phys_addr_t addr)
263 {
264 #if defined(CONFIG_IA64)
265 	/*
266 	 * On ia64, we ignore O_DSYNC because we cannot tolerate memory
267 	 * attribute aliases.
268 	 */
269 	return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
270 #elif defined(CONFIG_MIPS)
271 	{
272 		extern int __uncached_access(struct file *file,
273 					     unsigned long addr);
274 
275 		return __uncached_access(file, addr);
276 	}
277 #else
278 	/*
279 	 * Accessing memory above the top the kernel knows about or through a
280 	 * file pointer
281 	 * that was marked O_DSYNC will be done non-cached.
282 	 */
283 	if (file->f_flags & O_DSYNC)
284 		return 1;
285 	return addr >= __pa(high_memory);
286 #endif
287 }
288 #endif
289 
phys_mem_access_prot(struct file * file,unsigned long pfn,unsigned long size,pgprot_t vma_prot)290 static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
291 				     unsigned long size, pgprot_t vma_prot)
292 {
293 #ifdef pgprot_noncached
294 	phys_addr_t offset = pfn << PAGE_SHIFT;
295 
296 	if (uncached_access(file, offset))
297 		return pgprot_noncached(vma_prot);
298 #endif
299 	return vma_prot;
300 }
301 #endif
302 
303 #ifndef CONFIG_MMU
get_unmapped_area_mem(struct file * file,unsigned long addr,unsigned long len,unsigned long pgoff,unsigned long flags)304 static unsigned long get_unmapped_area_mem(struct file *file,
305 					   unsigned long addr,
306 					   unsigned long len,
307 					   unsigned long pgoff,
308 					   unsigned long flags)
309 {
310 	if (!valid_mmap_phys_addr_range(pgoff, len))
311 		return (unsigned long) -EINVAL;
312 	return pgoff << PAGE_SHIFT;
313 }
314 
315 /* can't do an in-place private mapping if there's no MMU */
private_mapping_ok(struct vm_area_struct * vma)316 static inline int private_mapping_ok(struct vm_area_struct *vma)
317 {
318 	return vma->vm_flags & VM_MAYSHARE;
319 }
320 #else
321 #define get_unmapped_area_mem	NULL
322 
private_mapping_ok(struct vm_area_struct * vma)323 static inline int private_mapping_ok(struct vm_area_struct *vma)
324 {
325 	return 1;
326 }
327 #endif
328 
329 static const struct vm_operations_struct mmap_mem_ops = {
330 #ifdef CONFIG_HAVE_IOREMAP_PROT
331 	.access = generic_access_phys
332 #endif
333 };
334 
mmap_mem(struct file * file,struct vm_area_struct * vma)335 static int mmap_mem(struct file *file, struct vm_area_struct *vma)
336 {
337 	size_t size = vma->vm_end - vma->vm_start;
338 	phys_addr_t offset = (phys_addr_t)vma->vm_pgoff << PAGE_SHIFT;
339 
340 	/* It's illegal to wrap around the end of the physical address space. */
341 	if (offset + (phys_addr_t)size - 1 < offset)
342 		return -EINVAL;
343 
344 	if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
345 		return -EINVAL;
346 
347 	if (!private_mapping_ok(vma))
348 		return -ENOSYS;
349 
350 	if (!range_is_allowed(vma->vm_pgoff, size))
351 		return -EPERM;
352 
353 	if (!phys_mem_access_prot_allowed(file, vma->vm_pgoff, size,
354 						&vma->vm_page_prot))
355 		return -EINVAL;
356 
357 	vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
358 						 size,
359 						 vma->vm_page_prot);
360 
361 	vma->vm_ops = &mmap_mem_ops;
362 
363 	/* Remap-pfn-range will mark the range VM_IO */
364 	if (remap_pfn_range(vma,
365 			    vma->vm_start,
366 			    vma->vm_pgoff,
367 			    size,
368 			    vma->vm_page_prot)) {
369 		return -EAGAIN;
370 	}
371 	return 0;
372 }
373 #endif	/* CONFIG_DEVMEM */
374 
375 #ifdef CONFIG_DEVKMEM
mmap_kmem(struct file * file,struct vm_area_struct * vma)376 static int mmap_kmem(struct file *file, struct vm_area_struct *vma)
377 {
378 	unsigned long pfn;
379 
380 	/* Turn a kernel-virtual address into a physical page frame */
381 	pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
382 
383 	/*
384 	 * RED-PEN: on some architectures there is more mapped memory than
385 	 * available in mem_map which pfn_valid checks for. Perhaps should add a
386 	 * new macro here.
387 	 *
388 	 * RED-PEN: vmalloc is not supported right now.
389 	 */
390 	if (!pfn_valid(pfn))
391 		return -EIO;
392 
393 	vma->vm_pgoff = pfn;
394 	return mmap_mem(file, vma);
395 }
396 #endif
397 
398 #ifdef CONFIG_DEVKMEM
399 /*
400  * This function reads the *virtual* memory as seen by the kernel.
401  */
read_kmem(struct file * file,char __user * buf,size_t count,loff_t * ppos)402 static ssize_t read_kmem(struct file *file, char __user *buf,
403 			 size_t count, loff_t *ppos)
404 {
405 	unsigned long p = *ppos;
406 	ssize_t low_count, read, sz;
407 	char *kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
408 	int err = 0;
409 
410 	read = 0;
411 	if (p < (unsigned long) high_memory) {
412 		low_count = count;
413 		if (count > (unsigned long)high_memory - p)
414 			low_count = (unsigned long)high_memory - p;
415 
416 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
417 		/* we don't have page 0 mapped on sparc and m68k.. */
418 		if (p < PAGE_SIZE && low_count > 0) {
419 			sz = size_inside_page(p, low_count);
420 			if (clear_user(buf, sz))
421 				return -EFAULT;
422 			buf += sz;
423 			p += sz;
424 			read += sz;
425 			low_count -= sz;
426 			count -= sz;
427 		}
428 #endif
429 		while (low_count > 0) {
430 			sz = size_inside_page(p, low_count);
431 
432 			/*
433 			 * On ia64 if a page has been mapped somewhere as
434 			 * uncached, then it must also be accessed uncached
435 			 * by the kernel or data corruption may occur
436 			 */
437 			kbuf = xlate_dev_kmem_ptr((char *)p);
438 
439 			if (copy_to_user(buf, kbuf, sz))
440 				return -EFAULT;
441 			buf += sz;
442 			p += sz;
443 			read += sz;
444 			low_count -= sz;
445 			count -= sz;
446 		}
447 	}
448 
449 	if (count > 0) {
450 		kbuf = (char *)__get_free_page(GFP_KERNEL);
451 		if (!kbuf)
452 			return -ENOMEM;
453 		while (count > 0) {
454 			sz = size_inside_page(p, count);
455 			if (!is_vmalloc_or_module_addr((void *)p)) {
456 				err = -ENXIO;
457 				break;
458 			}
459 			sz = vread(kbuf, (char *)p, sz);
460 			if (!sz)
461 				break;
462 			if (copy_to_user(buf, kbuf, sz)) {
463 				err = -EFAULT;
464 				break;
465 			}
466 			count -= sz;
467 			buf += sz;
468 			read += sz;
469 			p += sz;
470 		}
471 		free_page((unsigned long)kbuf);
472 	}
473 	*ppos = p;
474 	return read ? read : err;
475 }
476 
477 
do_write_kmem(unsigned long p,const char __user * buf,size_t count,loff_t * ppos)478 static ssize_t do_write_kmem(unsigned long p, const char __user *buf,
479 				size_t count, loff_t *ppos)
480 {
481 	ssize_t written, sz;
482 	unsigned long copied;
483 
484 	written = 0;
485 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
486 	/* we don't have page 0 mapped on sparc and m68k.. */
487 	if (p < PAGE_SIZE) {
488 		sz = size_inside_page(p, count);
489 		/* Hmm. Do something? */
490 		buf += sz;
491 		p += sz;
492 		count -= sz;
493 		written += sz;
494 	}
495 #endif
496 
497 	while (count > 0) {
498 		char *ptr;
499 
500 		sz = size_inside_page(p, count);
501 
502 		/*
503 		 * On ia64 if a page has been mapped somewhere as uncached, then
504 		 * it must also be accessed uncached by the kernel or data
505 		 * corruption may occur.
506 		 */
507 		ptr = xlate_dev_kmem_ptr((char *)p);
508 
509 		copied = copy_from_user(ptr, buf, sz);
510 		if (copied) {
511 			written += sz - copied;
512 			if (written)
513 				break;
514 			return -EFAULT;
515 		}
516 		buf += sz;
517 		p += sz;
518 		count -= sz;
519 		written += sz;
520 	}
521 
522 	*ppos += written;
523 	return written;
524 }
525 
526 /*
527  * This function writes to the *virtual* memory as seen by the kernel.
528  */
write_kmem(struct file * file,const char __user * buf,size_t count,loff_t * ppos)529 static ssize_t write_kmem(struct file *file, const char __user *buf,
530 			  size_t count, loff_t *ppos)
531 {
532 	unsigned long p = *ppos;
533 	ssize_t wrote = 0;
534 	ssize_t virtr = 0;
535 	char *kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
536 	int err = 0;
537 
538 	if (p < (unsigned long) high_memory) {
539 		unsigned long to_write = min_t(unsigned long, count,
540 					       (unsigned long)high_memory - p);
541 		wrote = do_write_kmem(p, buf, to_write, ppos);
542 		if (wrote != to_write)
543 			return wrote;
544 		p += wrote;
545 		buf += wrote;
546 		count -= wrote;
547 	}
548 
549 	if (count > 0) {
550 		kbuf = (char *)__get_free_page(GFP_KERNEL);
551 		if (!kbuf)
552 			return wrote ? wrote : -ENOMEM;
553 		while (count > 0) {
554 			unsigned long sz = size_inside_page(p, count);
555 			unsigned long n;
556 
557 			if (!is_vmalloc_or_module_addr((void *)p)) {
558 				err = -ENXIO;
559 				break;
560 			}
561 			n = copy_from_user(kbuf, buf, sz);
562 			if (n) {
563 				err = -EFAULT;
564 				break;
565 			}
566 			vwrite(kbuf, (char *)p, sz);
567 			count -= sz;
568 			buf += sz;
569 			virtr += sz;
570 			p += sz;
571 		}
572 		free_page((unsigned long)kbuf);
573 	}
574 
575 	*ppos = p;
576 	return virtr + wrote ? : err;
577 }
578 #endif
579 
580 #ifdef CONFIG_DEVPORT
read_port(struct file * file,char __user * buf,size_t count,loff_t * ppos)581 static ssize_t read_port(struct file *file, char __user *buf,
582 			 size_t count, loff_t *ppos)
583 {
584 	unsigned long i = *ppos;
585 	char __user *tmp = buf;
586 
587 	if (!access_ok(VERIFY_WRITE, buf, count))
588 		return -EFAULT;
589 	while (count-- > 0 && i < 65536) {
590 		if (__put_user(inb(i), tmp) < 0)
591 			return -EFAULT;
592 		i++;
593 		tmp++;
594 	}
595 	*ppos = i;
596 	return tmp-buf;
597 }
598 
write_port(struct file * file,const char __user * buf,size_t count,loff_t * ppos)599 static ssize_t write_port(struct file *file, const char __user *buf,
600 			  size_t count, loff_t *ppos)
601 {
602 	unsigned long i = *ppos;
603 	const char __user *tmp = buf;
604 
605 	if (!access_ok(VERIFY_READ, buf, count))
606 		return -EFAULT;
607 	while (count-- > 0 && i < 65536) {
608 		char c;
609 		if (__get_user(c, tmp)) {
610 			if (tmp > buf)
611 				break;
612 			return -EFAULT;
613 		}
614 		outb(c, i);
615 		i++;
616 		tmp++;
617 	}
618 	*ppos = i;
619 	return tmp-buf;
620 }
621 #endif
622 
read_null(struct file * file,char __user * buf,size_t count,loff_t * ppos)623 static ssize_t read_null(struct file *file, char __user *buf,
624 			 size_t count, loff_t *ppos)
625 {
626 	return 0;
627 }
628 
write_null(struct file * file,const char __user * buf,size_t count,loff_t * ppos)629 static ssize_t write_null(struct file *file, const char __user *buf,
630 			  size_t count, loff_t *ppos)
631 {
632 	return count;
633 }
634 
aio_read_null(struct kiocb * iocb,const struct iovec * iov,unsigned long nr_segs,loff_t pos)635 static ssize_t aio_read_null(struct kiocb *iocb, const struct iovec *iov,
636 			     unsigned long nr_segs, loff_t pos)
637 {
638 	return 0;
639 }
640 
aio_write_null(struct kiocb * iocb,const struct iovec * iov,unsigned long nr_segs,loff_t pos)641 static ssize_t aio_write_null(struct kiocb *iocb, const struct iovec *iov,
642 			      unsigned long nr_segs, loff_t pos)
643 {
644 	return iov_length(iov, nr_segs);
645 }
646 
pipe_to_null(struct pipe_inode_info * info,struct pipe_buffer * buf,struct splice_desc * sd)647 static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
648 			struct splice_desc *sd)
649 {
650 	return sd->len;
651 }
652 
splice_write_null(struct pipe_inode_info * pipe,struct file * out,loff_t * ppos,size_t len,unsigned int flags)653 static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out,
654 				 loff_t *ppos, size_t len, unsigned int flags)
655 {
656 	return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
657 }
658 
read_iter_zero(struct kiocb * iocb,struct iov_iter * iter)659 static ssize_t read_iter_zero(struct kiocb *iocb, struct iov_iter *iter)
660 {
661 	size_t written = 0;
662 
663 	while (iov_iter_count(iter)) {
664 		size_t chunk = iov_iter_count(iter), n;
665 		if (chunk > PAGE_SIZE)
666 			chunk = PAGE_SIZE;	/* Just for latency reasons */
667 		n = iov_iter_zero(chunk, iter);
668 		if (!n && iov_iter_count(iter))
669 			return written ? written : -EFAULT;
670 		written += n;
671 		if (signal_pending(current))
672 			return written ? written : -ERESTARTSYS;
673 		cond_resched();
674 	}
675 	return written;
676 }
677 
mmap_zero(struct file * file,struct vm_area_struct * vma)678 static int mmap_zero(struct file *file, struct vm_area_struct *vma)
679 {
680 #ifndef CONFIG_MMU
681 	return -ENOSYS;
682 #endif
683 	if (vma->vm_flags & VM_SHARED)
684 		return shmem_zero_setup(vma);
685 	return 0;
686 }
687 
write_full(struct file * file,const char __user * buf,size_t count,loff_t * ppos)688 static ssize_t write_full(struct file *file, const char __user *buf,
689 			  size_t count, loff_t *ppos)
690 {
691 	return -ENOSPC;
692 }
693 
694 /*
695  * Special lseek() function for /dev/null and /dev/zero.  Most notably, you
696  * can fopen() both devices with "a" now.  This was previously impossible.
697  * -- SRB.
698  */
null_lseek(struct file * file,loff_t offset,int orig)699 static loff_t null_lseek(struct file *file, loff_t offset, int orig)
700 {
701 	return file->f_pos = 0;
702 }
703 
704 #if defined(CONFIG_DEVMEM) || defined(CONFIG_DEVKMEM) || defined(CONFIG_DEVPORT)
705 
706 /*
707  * The memory devices use the full 32/64 bits of the offset, and so we cannot
708  * check against negative addresses: they are ok. The return value is weird,
709  * though, in that case (0).
710  *
711  * also note that seeking relative to the "end of file" isn't supported:
712  * it has no meaning, so it returns -EINVAL.
713  */
memory_lseek(struct file * file,loff_t offset,int orig)714 static loff_t memory_lseek(struct file *file, loff_t offset, int orig)
715 {
716 	loff_t ret;
717 
718 	mutex_lock(&file_inode(file)->i_mutex);
719 	switch (orig) {
720 	case SEEK_CUR:
721 		offset += file->f_pos;
722 	case SEEK_SET:
723 		/* to avoid userland mistaking f_pos=-9 as -EBADF=-9 */
724 		if (IS_ERR_VALUE((unsigned long long)offset)) {
725 			ret = -EOVERFLOW;
726 			break;
727 		}
728 		file->f_pos = offset;
729 		ret = file->f_pos;
730 		force_successful_syscall_return();
731 		break;
732 	default:
733 		ret = -EINVAL;
734 	}
735 	mutex_unlock(&file_inode(file)->i_mutex);
736 	return ret;
737 }
738 
739 #endif
740 
741 #if defined(CONFIG_DEVMEM) || defined(CONFIG_DEVKMEM) || defined(CONFIG_DEVPORT)
open_port(struct inode * inode,struct file * filp)742 static int open_port(struct inode *inode, struct file *filp)
743 {
744 	return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
745 }
746 #endif
747 
748 #define zero_lseek	null_lseek
749 #define full_lseek      null_lseek
750 #define write_zero	write_null
751 #define aio_write_zero	aio_write_null
752 #define open_mem	open_port
753 #define open_kmem	open_mem
754 
755 #ifdef CONFIG_DEVMEM
756 static const struct file_operations mem_fops = {
757 	.llseek		= memory_lseek,
758 	.read		= read_mem,
759 	.write		= write_mem,
760 	.mmap		= mmap_mem,
761 	.open		= open_mem,
762 	.get_unmapped_area = get_unmapped_area_mem,
763 };
764 #endif
765 
766 #ifdef CONFIG_DEVKMEM
767 static const struct file_operations kmem_fops = {
768 	.llseek		= memory_lseek,
769 	.read		= read_kmem,
770 	.write		= write_kmem,
771 	.mmap		= mmap_kmem,
772 	.open		= open_kmem,
773 	.get_unmapped_area = get_unmapped_area_mem,
774 };
775 #endif
776 
777 static const struct file_operations null_fops = {
778 	.llseek		= null_lseek,
779 	.read		= read_null,
780 	.write		= write_null,
781 	.aio_read	= aio_read_null,
782 	.aio_write	= aio_write_null,
783 	.splice_write	= splice_write_null,
784 };
785 
786 #ifdef CONFIG_DEVPORT
787 static const struct file_operations port_fops = {
788 	.llseek		= memory_lseek,
789 	.read		= read_port,
790 	.write		= write_port,
791 	.open		= open_port,
792 };
793 #endif
794 
795 static const struct file_operations zero_fops = {
796 	.llseek		= zero_lseek,
797 	.read		= new_sync_read,
798 	.write		= write_zero,
799 	.read_iter	= read_iter_zero,
800 	.aio_write	= aio_write_zero,
801 	.mmap		= mmap_zero,
802 };
803 
804 /*
805  * capabilities for /dev/zero
806  * - permits private mappings, "copies" are taken of the source of zeros
807  * - no writeback happens
808  */
809 static struct backing_dev_info zero_bdi = {
810 	.name		= "char/mem",
811 	.capabilities	= BDI_CAP_MAP_COPY | BDI_CAP_NO_ACCT_AND_WRITEBACK,
812 };
813 
814 static const struct file_operations full_fops = {
815 	.llseek		= full_lseek,
816 	.read		= new_sync_read,
817 	.read_iter	= read_iter_zero,
818 	.write		= write_full,
819 };
820 
821 static const struct memdev {
822 	const char *name;
823 	umode_t mode;
824 	const struct file_operations *fops;
825 	struct backing_dev_info *dev_info;
826 } devlist[] = {
827 #ifdef CONFIG_DEVMEM
828 	 [1] = { "mem", 0, &mem_fops, &directly_mappable_cdev_bdi },
829 #endif
830 #ifdef CONFIG_DEVKMEM
831 	 [2] = { "kmem", 0, &kmem_fops, &directly_mappable_cdev_bdi },
832 #endif
833 	 [3] = { "null", 0666, &null_fops, NULL },
834 #ifdef CONFIG_DEVPORT
835 	 [4] = { "port", 0, &port_fops, NULL },
836 #endif
837 	 [5] = { "zero", 0666, &zero_fops, &zero_bdi },
838 	 [7] = { "full", 0666, &full_fops, NULL },
839 	 [8] = { "random", 0666, &random_fops, NULL },
840 	 [9] = { "urandom", 0666, &urandom_fops, NULL },
841 #ifdef CONFIG_PRINTK
842 	[11] = { "kmsg", 0644, &kmsg_fops, NULL },
843 #endif
844 };
845 
memory_open(struct inode * inode,struct file * filp)846 static int memory_open(struct inode *inode, struct file *filp)
847 {
848 	int minor;
849 	const struct memdev *dev;
850 
851 	minor = iminor(inode);
852 	if (minor >= ARRAY_SIZE(devlist))
853 		return -ENXIO;
854 
855 	dev = &devlist[minor];
856 	if (!dev->fops)
857 		return -ENXIO;
858 
859 	filp->f_op = dev->fops;
860 	if (dev->dev_info)
861 		filp->f_mapping->backing_dev_info = dev->dev_info;
862 
863 	/* Is /dev/mem or /dev/kmem ? */
864 	if (dev->dev_info == &directly_mappable_cdev_bdi)
865 		filp->f_mode |= FMODE_UNSIGNED_OFFSET;
866 
867 	if (dev->fops->open)
868 		return dev->fops->open(inode, filp);
869 
870 	return 0;
871 }
872 
873 static const struct file_operations memory_fops = {
874 	.open = memory_open,
875 	.llseek = noop_llseek,
876 };
877 
mem_devnode(struct device * dev,umode_t * mode)878 static char *mem_devnode(struct device *dev, umode_t *mode)
879 {
880 	if (mode && devlist[MINOR(dev->devt)].mode)
881 		*mode = devlist[MINOR(dev->devt)].mode;
882 	return NULL;
883 }
884 
885 static struct class *mem_class;
886 
chr_dev_init(void)887 static int __init chr_dev_init(void)
888 {
889 	int minor;
890 	int err;
891 
892 	err = bdi_init(&zero_bdi);
893 	if (err)
894 		return err;
895 
896 	if (register_chrdev(MEM_MAJOR, "mem", &memory_fops))
897 		printk("unable to get major %d for memory devs\n", MEM_MAJOR);
898 
899 	mem_class = class_create(THIS_MODULE, "mem");
900 	if (IS_ERR(mem_class))
901 		return PTR_ERR(mem_class);
902 
903 	mem_class->devnode = mem_devnode;
904 	for (minor = 1; minor < ARRAY_SIZE(devlist); minor++) {
905 		if (!devlist[minor].name)
906 			continue;
907 
908 		/*
909 		 * Create /dev/port?
910 		 */
911 		if ((minor == DEVPORT_MINOR) && !arch_has_dev_port())
912 			continue;
913 
914 		device_create(mem_class, NULL, MKDEV(MEM_MAJOR, minor),
915 			      NULL, devlist[minor].name);
916 	}
917 
918 	return tty_init();
919 }
920 
921 fs_initcall(chr_dev_init);
922